Welcome to this week's edition of Rushing Robotics! If this week had a theme, it would be convergence — the moment when several long-running sci-fi promises stopped living in the "someday" file and started showing up as products, factories, and clinical results. Computers made of squishy polymers, DNA, and light. Robots with their own ambulance service and their own homes. A drug trial that measured age reversal with six independent clocks. Let's get into it.

🤯 Mind-Blowing

Computing left silicon behind this week — repeatedly. An MIT team built a brain-inspired device that thinks through physical movement, using a two-nanometer polymer film that remembers every voltage ever applied to it. A DNA computer floating in a droplet of salt water ran ten programs, including 100-bit addition, with no wires and no continuous power. Princeton engineers invented a semiconductor you can program with flashes of light, erasing and rewriting its behavior after it's built. And in Tokyo, the robot ecosystem matured in two very different directions: the city's first repair ambulance for humanoids hit the streets, while a startup began selling Iron Man-style homes with chore-doing robotic arms built into the ceiling.

🔊 Industry Insights & Updates

The industrial layer is scaling just as fast as the lab. UBTECH fired up a purpose-built humanoid factory in Liuzhou rated for 10,000 units a year, going from 1,000-unit to 10,000-unit capacity in under nine months. BYD brought a 1.5-megawatt-charging electric truck to Europe that adds 60 percent battery in 20 minutes — an answer to trucking's downtime problem. Agility debuted Digit 5, a warehouse humanoid shaped by over 65,000 hours of field operation with $300 million in conditional commitments behind it. And the IAEA raised its nuclear forecast for the sixth straight year, projecting up to 946 small modular reactors by 2060 — a quiet energy backbone for everything above.

🧬 BioTech

Longevity science delivered what may be its biggest week ever. Insilico confirmed biological age reversal in a human trial — unanimously, across six independent proteomic aging clocks — using an AI-designed drug, with the strongest group showing three to six years of age reduction. A Stanford team simulated an entire drug company with AI agents and gained real-world validation for its theory of why some drug targets win and others fail. And a multi-marker blood test captured pancreatic cancer's earliest warning signs in high-risk patients, potentially shifting detection to a stage where intervention is still possible.

💡 Products/Tools of the Week

Three tools worth your time: ArchOne AI turns room photos and sketches into comparable architectural design directions, YuE2 lets you inspect and edit an AI song’s melody and chords before it ever becomes audio, and CaptureNote converts screenshots of slides and diagrams into structured, searchable notes — with a fully offline iPhone option.

🎥 Video Section

This week's watchlist takes you inside the machines: launch footage of Agility's Digit 5 in action, a Munro Live teardown of the Unitree G1's actuators to see exactly how these humanoids move, and a look at IFA 2026, where robots crossed some genuinely uncanny thresholds.

Weeks like this one make the future feel less like a destination and more like something being assembled in real time — a DNA computer here, a humanoid factory there, a measured reversal of human aging in a clinical trial. The distance between laboratory curiosity and deployed infrastructure keeps collapsing. If six aging clocks can agree that a person grew younger, and a droplet of salt water can do arithmetic, imagine what the next five years will argue is impossible. Stay hungry, stay futurish!

🤯 Mind-Blowing

• Brain-inspired computing took a small, squishy step forward with a new nanoscale device that thinks through physical movement rather than shuttling data between separate memory and processor units. An MIT team designed the platform to replicate how neurons accumulate signals until they reach a firing threshold, using soft polymers that compress and bounce back under voltage. The breakthrough rests on a viscoelastic polymer film just two nanometers thick placed between two electrodes; it prevents the surfaces from sticking together permanently — a barrier that had crippled earlier attempts at nanoscale mechanical computers — while its gradual rebound records the history of everything applied to it. The team's chip contains hundreds of these devices, and by building computation functions directly into the material itself, the platform needs no extra capacitors or complex circuitry, promising high energy efficiency in an ultracompact footprint. Applications range from interactive medical patches and smart robotics to adaptive environmental monitoring, and the team now plans to integrate sensing alongside memory and computing for fully adaptive nanomechanical systems.

• A computer made of nothing but DNA strands floating in a droplet of salt water has executed multiplication, division and addition — math once reserved for silicon chips — without a single wire or continuous power supply. An MU team built what is called a Scaffolded DNA Computer, mixing short DNA pieces with a long scaffold strand inside a small droplet, then kick-starting the reaction with a brief pulse of heat. As the liquid cools, billions to trillions of strands drift together, react and self-assemble into molecular patterns that are the answer. The droplet ran ten distinct programs: a simple 10 + 3 computation finished in just 30 seconds, while complex 100-bit additions of numbers up to 34 million took roughly 14 hours — still the fastest non-trivial programs in the DNA computing literature. The system proved robust and renewable, completing up to 25 consecutive calculations in the same droplet with no performance loss. Future versions could archive data for millennia or even travel through bloodstreams diagnosing disease cell by cell.

• New hardware that thinks of light as its programming language took shape in a New Jersey lab, pointing toward sensors and computers whose jobs can change after they are built. A Princeton team layered light-responsive molecules onto ultra-thin 2D semiconductors so that each flash of a chosen wavelength reshapes the molecules and rewrites the electrical and optical behavior beneath them. Picture inkblots forming and dissolving, pixel by pixel, into configurations a device can use. The result is a material inspired by living systems: instead of a fixed state, it senses and responds to its surroundings continuously, a property the authors call essential for any truly smart material. Unlike standard switches, its states aren't merely on or off — its response slides along a gradual scale and can be reversed, enabling much finer control. The team has already demonstrated uniform one-inch-square films and working arrays of programmable switches, with circuits as the clear next step. Longer term, the work targets energy-efficient edge computing and optoelectronics that reconfigure themselves on demand.

• Emergency roadside service arrived for humanoid robots on September 8, when a Tokyo-based robotics firm rolled out the city's first dedicated repair ambulance. GMO Air developed the vehicle as a proof of concept, reflecting the kind of support ecosystem that long ago formed around cars, aircraft and industrial machinery — and that robots will need as they take on commercially critical jobs. Rather than autonomous self-repair machines, the service pairs old-school human technicians with new-age clientele: engineers drive the van, perform on-site diagnosis, and either fix the unit or swap it for the spare riding in back. Broken robots then head to a repair lab while their replacements keep working, compressing what was once a multi-day service interruption into a same-day visit. Response times of one to two hours are targeted across the Tokyo area, including overnight calls, and the service links to a 70,000-yen-monthly subscription covering replacement units.

• An Iron Man-style house robot moved one step from fiction to fixtures when a Tokyo startup unveiled arms that do chores from the ceiling down. MW drew on a fictional AI butler for its vision, but its business plan is grounded in high-end real estate: more than 1,000 applications are already in hand for robot-integrated homes, with pricing between roughly $1.3 million and $2.7 million. The concept rests on three pillars — custom residential architecture, integrated hardware called the MW bot, and a dedicated software operating system. The pitch is that fixed machines beat mobile humanoids on safety, space and efficiency: rail-mounted arms can also be contained more easily if something goes wrong. The first-generation robot is already integrated into a trailer home, billed as the first fusion of architecture and AI robotics, and the company plans to demonstrate full autonomy by September 2027. Sales of robot-ready houses begin with Tokyo in 2028, international expansion follows from 2029, and future build sites include the coastal city of Futtsu and the resort town of Karuizawa.

🔊 Industry Insights & Updates

• Mass production of humanoid robots crossed from prototype workshops into automotive-grade industrialization this week in southern China. UBTECH commissioned its Liuzhou smart factory as the first plant purpose-built for a 10,000-plus annual output of embodied-intelligence humanoids, running a ten-minute takt time validated in simulation before construction began. The leap was fast: the thousandth Walker S2 rolled off the line only in December 2025, meaning the jump from 1,000-unit to 10,000-unit capacity took under nine months. A proprietary "smart brain" system orchestrates scheduling and operations across the floor, where robots act as both the product and the workforce — a closed loop from raw material intake to final shipping. Analysts nonetheless caution that announced capacity is not output: real performance hinges on industrial orders, component supply and manufacturing economics as rival humanoid programs accelerate worldwide.

• Long-haul trucking's electrification problem — downtime — got a direct answer at Europe's biggest commercial vehicle fair. BYD unveiled its most complete zero-emission portfolio yet on September 14, anchored by a flagship tractor whose 1.5-megawatt charging takes the battery from 20 to 80 percent in around 20 minutes, three times faster than the same session on a standard DC connection. The pitch goes beyond a single truck: a Total Cost of Mobility philosophy bundles vehicles, energy consumption, electricity pricing, charging infrastructure, lifecycle management and financing into one vertically integrated ecosystem spanning trucks, Blade Batteries, megawatt chargers, storage and energy management. With its commercial vehicles already operating in more than 100 countries and over 400 cities, the company is positioning itself not just as a truck maker but as a one-stop provider of the energy infrastructure fleets need to run profitably. European arrival is expected in 2027, with a five-star Euro NCAP safety credential adding credibility to the market push.

• Warehouse automation crossed a milestone this week with the debut of a humanoid designed from the ground up for "cooperatively safe" work next to human colleagues. Agility unveiled Digit 5, a machine shaped by more than 65,000 hours of prior field operation, featuring new bending legs that let it crouch, kneel and collect cargo without the awkward gait of its predecessor. The robot reaches 7.2 feet high yet still fits through standard doorways, lifts up to 50 pounds repeatedly, and returns to full duty after a nine-minute charge — enough for over 20 productive hours in a 24-hour day. Its swappable grippers handle everything from depalletizing to machine tending, while an independent safety controller manages how it behaves around people. The company reports more than $300 million in conditional commitments, and production will run at a facility with capacity for up to 10,000 robots annually, with commercial rollout expected in 2027.

• Nearly 1,000 small modular reactors could be powering the planet by 2060 under a fresh long-range look at nuclear energy, the first to stretch projections all the way to that year. The IAEA raised its global nuclear forecast for the sixth consecutive year in its report Energy, Electricity and Nuclear Power Estimates for the Period up to 2060. At the end of 2025, 413 reactors provided 377.1 GWe worldwide; under the high-case scenario that capacity would hit 1,045 GWe by 2050 and 1,284 GWe by 2060 — 3.4 times today's level. SMRs, defined as reactors of up to 300 MWe and with more than 100 designs now in development, could supply 28% of the 1,017 GWe in new capacity, or 284 GWe — equivalent to 946 units averaging 300 MWe, or 1,136 at 250 MWe. Regional uptake varies: North America could source 60% of new capacity from SMRs, with 40% shares in South-Eastern Asia and Latin America. Even the low case — 400 to 480 SMRs — reflects momentum from energy-security concerns since 2022 and renewed backing from international financial institutions.

Subscribe now

🧬 BioTech

• Biological age reversal has now been confirmed in a human clinical trial — unanimously so, by six independent measurement tools — a first for the longevity field. Insilico presented the findings at an international AI-in-healthcare conference in Paris this week, following their publication in Nature Biotechnology. The analysis drew on a 12-week Phase IIa trial of an AI-designed drug called rentosertib in 42 patients with idiopathic pulmonary fibrosis, lung scarring that affects roughly 5 million people worldwide. Researchers tracked 2,841 proteins in patient blood samples and ran the data through six proteomic aging clocks built separately by different research groups with different methods. Every clock agreed: treated patients ended up biologically younger than those on placebo. The strongest signal appeared at week 4 in the 30 mg twice-daily group, showing an average age reduction of three to four years and up to six on one clock. Under the hood, the drug works as a senomorphic agent, suppressing proteins tied to cellular senescence and growth-factor pathways known to accelerate aging. All trial proteomic data and analysis code were released publicly, setting a template for embedding aging endpoints into future drug trials.

• An entire drug company was simulated with AI agents, and its predictions about what makes medicines succeed just gained independent, real-world validation. A Stanford team designed the system to answer a deceptively simple question: what hidden biological features separate winning drug candidates from failures? The answer concerns "bimodality" — whether a targeted gene behaves more like a light switch than a dimmer dial — combined with cell-type specificity. The theory is straightforward: an on-off target confined to one cell type may simply be easier and safer to control than one with a broad spectrum of activity. Beyond the statistics, the system's lung cancer proposal traced a biological mechanism in which cancer-associated connective tissue cells express B7-H3 to suppress nearby immune cells, effectively cloaking tumors — then designed a payload-delivering therapy around it. The researchers now stress that humans, physical experimentation and validation remain the indispensable conduit through which AI makes its impact, and they're not pursuing B7-H3 themselves since another firm already shepherds it toward the clinic.

• Pancreatic cancer's earliest warning signs were captured in a simple blood draw, according to a large international study out this week. The experimental test merges several biological signals into one score because, as the researchers note, "most biomarkers tell you one part of the story" — combining them gives a clearer picture of what may be happening inside the pancreas. What sets the approach apart from previous blood-based attempts is both the multi-marker design and the choice of who it was tested on: people with familial or inherited genetic risk, pancreatic cysts and chronic pancreatitis, not just healthy controls. Current monitoring for these individuals relies on imaging that is invasive, expensive and only partly helpful at spotting the earliest tumors. A reliable blood test could instead triage who actually needs a closer look — and the team frames the stakes plainly: a stage shift is not just a statistic, since the earlier pancreatic cancer is found, the greater the chance that meaningful intervention remains possible.

  ---

💡Products/tools of the week

• A new AI-powered spatial design visualization platform has arrived for early-stage concept work. ArchOne AI lets architects, interior designers, landscape designers, small studios, consultants, students, and homeowners upload room photos, facade images, outdoor site shots, sketches, renders, or reference images to generate and compare visual design directions. The platform covers architecture, interiors, exteriors, and landscapes, supporting material palettes, lighting mood, furniture direction, facade updates, and planting concepts. ArchOne runs on a freemium model: users receive free credits at sign-up, with paid plans and credit packs available for heavier usage.

• A new AI-powered spatial design visualization platform has arrived for early-stage concept work. ArchOne AI lets architects, interior designers, landscape designers, small studios, consultants, students, and homeowners upload room photos, facade images, outdoor site shots, sketches, renders, or reference images to generate and compare visual design directions. The platform covers architecture, interiors, exteriors, and landscapes, supporting material palettes, lighting mood, furniture direction, facade updates, and planting concepts. ArchOne runs on a freemium model: users receive free credits at sign-up, with paid plans and credit packs available for heavier usage.

• AI-made music just became inspectable and editable before it becomes sound. YuE2, an open-source system from Multimodal Art Projection and HKUST released this month, unifies symbolic planning with audio synthesis: it turns lyrics and a style prompt into an ABC-notation score, lets users adjust melody, chords, and structure, and then renders a complete song with vocals and accompaniment. Targeting frontier-quality output across languages and musical styles, the model offers full melody-and-chords planning, melody-only mode, or direct generation with no plan at all, and is already available for local runs including a ComfyUI workflow.

• Screenshot-to-notes capture just moved beyond plain OCR with CaptureNote, a tool that turns images of lecture slides, textbook pages, charts, and diagrams into fully structured notes. Built by the CaptureNote team (CN), the app uses multimodal AI to read visual hierarchy, interpret diagrams, and understand relationships between elements rather than simply extracting raw text. Each upload produces a note with a title, summary, key points, and key terms, all saved automatically to a private searchable library. Students and researchers can organize notes into folders, search them by content, edit them after capture, and export everything as Markdown for use in other tools. An iPhone app extends the workflow to mobile, offering offline, on-device processing as a one-time purchase, so sensitive study material never has to leave the user's device.

  ---

• Robot Harbour: Agility Digit 5: Humanoid Robot Launch Footage | Robot Harbour

• Munro Live: Unitree G1 Actuator Teardown: How This Humanoid Robot Moves

• The AI Nexus: IFA 2026 Revealed How Shockingly Human Robots Are Becoming

Welcome to this week's edition of Rushing Robotics! Time misbehaved this week. A drug wound biological clocks backward by years. A swarm of ten thousand AI agents solved in 88 hours a math problem that stumped humanity for 90. A robot finished each pollinated flower in under ten seconds flat. And in one operating room, a machine repaired a human heart without a single human hand in the loop. The future isn't arriving on schedule — it's early.

🤯 Mind-Blowing

A robot repaired a human heart entirely on its own, a first in medicine. Rentosertib reversed biological age across six independent aging clocks. A two-armed humanoid took over crop pollination. OpenAI's 10,000-agent system claims a Millennium Prize proof, and two researchers say it was built on their work. Plus, a CXL architecture that makes a whole data center behave like one computer.

🔊 Industry Insights & Updates

AEON humanoids entered Schaeffler's training gym ahead of a planned 1,000-robot fleet. Huawei published hard numbers showing stacked chips can be 55% denser and run cooler. Europe inaugurated the 60-petaflop Arrhenius supercomputer in Sweden. And engineers built a working laser out of deliberate imperfection.

🧬 BioTech

ECG-CLIP reads heart rhythms like a cardiologist after training on 1.7 million ECGs. Blocking one molecule reversed pulmonary arterial hypertension in preclinical models. And the immune attack behind Alzheimer's-like damage turns out to start outside the brain, which means existing drugs might be repurposed against it.

💡 Products/Tools of the Week

Describe a game in plain English and play it minutes later. Chat directly with your webpages. Polish rough screen recordings into studio demos in about 60 seconds. And let an AI watch the open web for you.

🎥 Video Section

On screen this week — Hong Kong's first robot-run convenience stores, fully autonomous humanoid combat from Unitree, a hard look at China's robot bubble, and AGIBOT's reinforcement-learning showcase.

What stands out this week is how much expertise is being compressed into machines and molecules — skills that took surgeons decades, instincts that took breeders generations, mathematics that took a century. That compression is accelerating, and the places it lands next are getting harder to predict and more interesting to watch. We'll keep tracking it, one week at a time. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A robot has seen a heart defect and fixed it by itself — a world first. A medical team used an AI-driven, ultrasound-guided robot to perform a transcatheter closure for congenital heart disease in a 48-year-old man, sealing the defect through a catheter without opening the chest. What distinguishes the system from surgeon-operated robots like da Vinci is its closed loop: deep learning links real-time ultrasound image recognition directly to physical motor control, so the robot autonomously scanned the cardiac tissue, mapped the lesion, and deployed the cardiac occluder with sub-millimeter precision rather than waiting for human commands. Because such procedures normally rely on X-ray fluoroscopy, switching to ultrasound also eliminates radiation exposure for patient and staff. The hospital says the technology could reduce human error, shorten procedure times, and offload expertise that usually takes decades to acquire — potentially scaling advanced cardiac care to ordinary clinics, battlefield triage, and remote medicine.

• Six independent clocks have agreed on something startling: a drug made patients look biologically younger. A Nature Biotechnology study reports that rentosertib — the first drug candidate with both an AI-discovered target and an AI-designed molecule — lowered predicted biological age in idiopathic pulmonary fibrosis patients according to all six proteomic aging clocks applied to their blood. Analyzing 12 weeks of data covering 2,841 proteins from 42 Phase IIa participants; the six clocks were built independently by Harvard, Oxford, PKU, and Insilico groups, yet every one showed a reversal trend in treated arms versus placebo. The peak came at Week 4 on 30 mg twice daily — roughly 3–4 years of reversal, up to 6 on one clock — and lung function tracked the same direction. Notably, the best age-reversal dose differed from the best lung-function dose, suggesting the geroprotective effect is partly independent of treating the disease; mechanistically, the drug suppressed senescence-driving proteins. Nobel laureate Michael Levitt called the cross-model agreement convincing but stressed the trial cannot yet separate slower aging from a treated lung. Rentosertib is now in Phase III trials in China, and the team published its data and code openly.

• Farm work just got its first two-armed specialist. GEAIR 2.0, a humanoid robot built to automate crop breeding, was unveiled at the 33rd China Beijing Seed Industry Conference, where it demonstrated fully autonomous hybrid pollination of tomato plants. Developed by a team at the Institute of Genetics and Developmental Biology under the Chinese Academy of Sciences, the robot swaps its predecessor's single flexible arm for dual arms that push through leaves and branches, and its accuracy at recognizing a flower's stigma — the needle-sized female organ — has risen from 85% to 92.8%. At the launch it held a pollen tube in one hand and a pollination brush in the other, finishing each pollination in under 10 seconds, with one dip of pollen covering roughly 50 flowers. The system works because the crops are engineered for the robot as much as the robot for the crops: gene editing creates male-sterile plants with exposed stigmas, a "crop-robot co-design" strategy published in a 2025 Cell study that matched manual pollination efficiency in tomato and has since been extended to soybean. Its developers say robots that work long hours in heat or cold could cut breeding labor and costs — a priority as China, fresh off a record 715-million-ton grain harvest in 2025, targets roughly 725 million tons of capacity by 2030.

• A 90-year-old mathematical mystery may have been solved by ten thousand artificial minds — and the triumph arrived handcuffed to a scandal. OpenAI says an internal system, peaking at roughly 10,000 coordinating agents, produced a Lean-formalized proof that the three-dimensional Navier–Stokes equations can develop a finite-time singularity: a vortex that collapses inward until velocity becomes unbounded while total energy stays finite, potentially resolving one of the seven Millennium Prize Problems. The campaign began September 1, took about 88 hours and 2.7 million messages, and GPT-6 Astra spent 17 more hours checking the proof in Lean; OpenAI is declining the $1 million prize and has released everything for outside review. Within hours, however, NYU mathematician Tristan Buckmaster and Anthropic researcher Levent Alpöge alleged the company may have built on their unpublished work. The pair had spent nearly a year on closely related fluid equations, partly inside OpenAI's Codex tool; Buckmaster says he learned on September 3 that details of their progress had reached OpenAI, and alleges OpenAI's Sébastien Bubeck proposed a joint announcement with Alpöge's credit removed — and, when Buckmaster threatened to go public, asked "Why would you ruin your career?" OpenAI calls the allegations "categorically false," saying no one saw the pair's work before release and no specific user data was accessed, while conceding it cannot fully rule out de-identified product data improving its models.

• Panmnesia and Meta have proposed a CXL-based architecture designed to make AI data centers behave like a single giant computer, tying CPUs, accelerators, and memory together across rack boundaries that today force traffic onto conventional networks with unpredictable delays — a growing problem for trillion-parameter models, where one lagging device stalls thousands of others. The design extends a Compute Express Link coherence domain across the facility using three core components — a high-fan-out non-blocking CXL switch, a link acceleration unit, and a fabric controller — plus optical links that overcome the distance limits of electrical signaling. In the reference configuration, one CPU coordinates 16 accelerators instead of 2, a single coherence domain spans up to 960 accelerators, and round-trip latency falls from microseconds to several hundred nanoseconds; failures also shrink from server-level to device-level.

🔊 Industry Insights & Updates

• A thousand-robot plan now has a training camp. AEON, the humanoid robot built by Hexagon Robotics, has entered Schaeffler's Humanoid Gym in Germany — a dedicated industrial facility where it will acquire manufacturing skills before any contact with a live production line. Hexagon Robotics and motion technology company Schaeffler announced the milestone as the next stage of a collaboration targeting the deployment of at least 1,000 AEON humanoids in the coming years. The gym runs on a train–validate–deploy model: every capability is first taught to the robot, then assessed, and only then prepared for real factory use. The arrangement works both ways, since Schaeffler simultaneously builds in-house expertise in operating, training, and integrating humanoids — a thousand-robot fleet needs people who can manage it.

• A 55% jump in transistor density without extra heat has challenged one of the biggest doubts about stacked chips. New measurements published by Huawei show its coming Kirin processor — the first built on the company's Tau Scaling Law and LogicFolding architecture — packs 55% more transistors per square millimetre than the Kirin 9030 Pro while cutting power consumption by up to 66% on key workloads and still running cooler than its predecessor. The research, directly rebuts warnings that vertically stacking logic would create a thermal bottleneck: by folding circuitry into stacked layers, signals travel shorter distances and lose less energy, while thermal planning routes heat away from the hottest modules. The stakes reach beyond one smartphone chip. Developed under sanctions that block China's access to advanced EUV lithography, the Tau approach claims a path to 1.4-nanometre-equivalent density by 2031, and Huawei has now published three papers on it since May.

• Sixty million billion calculations per second now hum away in Linköping. The EuroHPC Joint Undertaking and NAISS have inaugurated Arrhenius, a mid-range supercomputer delivering more than 60 petaflops, ranking 42nd on the June 2026 TOP500 list and 24th on the Green500 with over 61 billion operations per watt. The €68.5 million system is now open to researchers, companies, and public agencies across Europe for climate modeling, materials research, drug discovery, life sciences, and AI workloads too large for conventional systems — including projects involving sensitive personal data, which the machine is specifically secured to handle. Some researchers had already gained early access through the EuroHPC AI for Science call before the ceremony. Strategically, Arrhenius becomes the thirteenth supercomputer procured by EuroHPC as Europe races to control its own computing capacity, alongside 19 AI Factories and a planned quantum-computing layer.

• Imperfection has been engineered into a laser — and it works. Researchers at the University of Illinois Urbana-Champaign demonstrated a semiconductor laser that deliberately abandons the strict repeating patterns of conventional photonic-crystal surface-emitting lasers, replacing them with a quasi-periodic structure that still produced single-mode lasing at room temperature at 1.5 micrometers, the wavelength band where optical fibers lose the least signal. The key is the UIUC team's buried-dielectric platform: instead of etching fragile patterns into the semiconductor itself — where regrowth typically distorts them — the researchers patterned a silicon dioxide layer and buried it under epitaxial material, so what engineers draw is what they get. That freedom could let designers mix different photonic patterns on the same substrate, pointing toward more tunable, reliable lasers for sensing, communications, aerospace, and silicon-photonics lidar. The team, including engineering professor Kent Choquette and PhD candidate Erin Raftery, is candid about the limits: the device is a photopumped proof of concept, not yet electrically injected, and hasn't beaten conventional PCSELs on every measure.

Subscribe now

🧬 BioTech

• AI has learned to read an ECG the way a cardiologist does — broadly first, specifically later. Researchers have unveiled ECG-CLIP, a foundation model trained on over 1.7 million electrocardiograms and the clinician notes accompanying them, which the team says markedly improves detection and prediction of heart disease while slashing the labeled data such systems normally require. The model outperformed standard deep-learning, linear, and existing ECG foundation models across every task tested: spotting heart attacks, cardiac amyloidosis, and hypertrophic cardiomyopathy in an external set of more than 800,000 ECGs; forecasting atrial fibrillation from normal-rhythm recordings; and estimating 30-day survival and three-year risk of chronic kidney disease and type II diabetes.

• A single molecule's silencing has reversed a deadly lung disease, at least in animals. Researchers led by Yassine Sassi at the Fralin Biomedical Research Institute at VTC blocked microRNA-224 and reversed pulmonary arterial hypertension in preclinical models, using both a drug-like therapy and a gene therapy approach. The molecule, found at elevated levels in the lungs of PAH patients, drives abnormal blood vessel growth; blocking it improved vessel function, heart performance, and survival. Published in Science Translational Medicine, the work targets a disease that can kill within two to three years of diagnosis, and one for which current treatments only ease symptoms. Human trials are the next hurdle.

• The immune attack behind Alzheimer's-like damage starts outside the brain. WashU Medicine researchers found in mice that dendritic cells in lymph nodes — especially in the neck — prime T cells to invade the brain and drive tau-linked neurodegeneration. Eliminating those dendritic cells erased the brain's T-cell surge, sharply reduced damage, and preserved cognition, even with tau tangles unchanged. Because the priming happens beyond the blood-brain barrier, existing T-cell-modulating drugs could potentially be repurposed.

  ---

💡Products/tools of the week

• Playable web games can now be created just by describing an idea in plain English. Crayon AI, an AI-powered game engine, generates playable 2D or 3D games from a simple text prompt without requiring any coding or engine knowledge. Users type a genre, world, character, or concept, and Crayon builds the game. Finished games can be published to the Crayon Arcade or shared via a direct link, with users retaining full ownership of their code and IP. The platform is free during beta.

• Chat directly with webpages. That's what Tabbit Browser enables as an AI-native tool built to understand user context across browsing sessions. Users can automate tasks with a built-in Agent, create custom Skills for recurring workflows, and organize tabs for export to Notion or Obsidian. The browser also surfaces key podcast quotes, extracts transcripts, finds substantive comments, and triages content feeds by relevance.

• Rough screen recordings become polished product demos in about 60 seconds. DemoPolish, an AI-powered tool, lets users upload any MP4 from Loom, OBS, QuickTime, or similar platforms, then automatically rewrites the narration script and applies a studio-quality AI voiceover. Founders can choose from six AI voices to lock in a consistent brand voice across multiple clips. Pricing starts with a $1 three-day trial covering five videos, then shifts to $29 per month for 100 videos.

• Continuous monitoring of the open web is now automated. Skymel, a persistent web intelligence platform, scans hundreds of sources including social media, job boards, and regulatory filings. It uses AI to resolve entities and build a live knowledge graph by connecting fragmented signals. Skymel delivers results via webhooks, APIs, or synthesized reports, and integrates with tools like Salesforce and Snowflake to combine external and internal data.

  ---

• South China Morning Post: Hong Kong’s first robot-run convenience stores

• Unitree Robotics: Unitree Breakthrough: The World’s First Fully Autonomous Humanoid Robot Combat

• Wall Street Millennial: China’s Humanoid Robot Bubble Reaches Point Of Absurdity

• AGIBOT: AGIBOT WORLD 2026 Theme 3: Reinforcement Learning

Welcome to this week's edition of Rushing Robotics! Some weeks produce news; this one produced records. The largest functional map of human immunity ever assembled. The fastest quantum processor IBM has ever shipped. The first photonic chips to be flexible and transparent at the same time. The most capable AI model ever released. Records, of course, are just receipts — what matters is what they unlock. Let's see what this week's are worth.

🤯 Mind-Blowing

This week's stack of wonders runs from the microscopic to the era-defining. We start inside your own blood, where 33.4 million immune cells just revealed the hidden genetic switches governing health and disease — the largest contribution yet to the Billion Cells Project. Then: a 3D-printed speaker that beams private sound to a single airplane seat, a one-pot CRISPR test that brings lab-grade diagnostics into the home in 30 minutes, and flexible, transparent photonic chips straight out of MIT. And of course, the launch that dominated every feed on the planet — GPT-6 Astra takes the throne.

🔊 Industry Insights & Updates

Meanwhile, the deep-tech engine room never stopped humming. IBM's Nighthawk r2 quantum processor now resets its qubits in a microsecond blink — 25 times faster than its predecessor — while Chinese researchers opened a "fast lane" for quantum memory with 98% routing efficiency. In energy, Japan shortlisted a stellarator startup for its $400 million fusion program, and 3D-printed parts are officially heading into nuclear reactors.

🧬 BioTech

Biology had a quietly spectacular week. An arthritis pill regrew full heads of hair in patients with severe alopecia, three Seattle powerhouses launched an AI accelerator to design biology that evolution never tried, and a decade-long heart trial opened the door to life-saving implants for an entirely new group of patients.

💡 Products/Tools of the Week

This week's picks run from genuinely useful to delightfully unhinged: viral-hook radar for marketers, an AI video channel programmed by its own audience, one-click text editing inside images, and a single API for the world's academic papers.

🎥 Video Section

Three videos this week circle the same question: how close is the personal humanoid? Zeroth's answer costs $4,000 and stands in your living room, a student team shows how simulated motion becomes real walking, and Galileo X sketches what robots look like when form stops being a constraint.

Precision stacks — each new tool for controlling genes, light, sound, and machines becomes the foundation the next breakthrough stands on. Better still, much of this week's work is open: open datasets, open science, open platforms, so the next leap can come from anyone, anywhere. That's what makes this moment so thrilling to cover. Stay hungry, stay futurish!

🤯 Mind-Blowing

• Immunity now has a wiring diagram. Scientists have unveiled a massive, high-resolution functional map of human immune cells that promises to transform understanding of how genetics control health and disease. The team systematically stress-tested genes across the genome in 22 million human immune cells, using a technology called Perturb-seq to turn off nearly 12,800 different genes one by one in T cells taken from real blood donors rather than laboratory cell lines, so they could watch how genes work in their natural state. The result moves biology beyond cataloging DNA toward decoding the dynamic circuits that govern how genes behave, with the data showing these circuits operate very differently depending on whether cells are resting or fighting an infection, context the researchers say is critical for training AI models to predict cell behavior. The dataset is the largest contribution yet to Biohub’s Billion Cells Project, an open-source effort to build a one-billion-cell foundation for AI-accelerated biology, and was enabled by screening 33.4 million cells through partnerships with 10x Genomics and Ultima Genomics. The researchers, who have now pivoted with Weill Cancer Hub West toward mapping how genetic changes affect T cells inside tumors, say the map offers a new framework for designing cancer immunotherapies and treating autoimmune conditions, and could become a standard lookup table for the entire field.

• Private sound is no longer science fiction. A new kind of speaker can beam audio to a single listener without headphones, and researchers) designed the low-cost, single-element device, called MiPAL, to do it with acoustic and elastic metamaterials. The system works as a parametric array loudspeaker, modulating ultrasound waves with audible content so the nonlinear behavior of sound in air focuses audio like a ray of light, while a dual-metamaterial safeguard built into one transducer solves the problems that have hobbled simpler designs: an acoustic metasurface at the front acts as a lens, shaping ultrasonic waves into a narrow beam, and elastic meta-units at the rear suppress parasitic vibrations that would otherwise leak sound sideways. The device directs audio across 500 Hz to 10 kHz, more than four octaves of the audible spectrum, and in simulated aircraft seating tests using real broadcast and music signals, it delivered isolated sound to individual passengers without disturbing neighboring seats. Because its modular metamaterial structures are 3D printed, engineers can quickly adapt the design to different seat spacings, beam angles or operating conditions.

• A 30-minute home test now matches the lab. HKUST researchers, jointly with the Chinese University of Hong Kong, developed TEMPO, a "one-pot" CRISPR platform that folds an entire nucleic acid test into a single reaction tube, using thermodynamically engineered DNA primers to make amplification and detection happen automatically in the right order. The platform reached attomolar sensitivity in a single-step 30-minute workflow, proved concordant with PCR on HIV, influenza A and SARS-CoV-2 samples, and needs no cold chain or bulky instruments, suiting homes, clinics, airports and remote settings. Published in Nature Communications, the work could bring lab-grade testing for viruses and hereditary conditions into the home, with a sample-to-answer device planned within the next year.

• Light-based chips just learned to bend. MIT researchers have developed a wafer-scale process that makes silicon-photonics chips, which move data using light rather than electrical signals, both flexible and transparent for the first time, a combination that had previously eluded scalable manufacturing. Working with engineers at NY Creates’ Albany NanoTech Complex, the team produced 300-millimeter wafers carrying chips only a few microns thick, using stable foundry fabrication tools that give them confidence large numbers of devices will perform to specification. In tests, a chip survived thousands of bends around cylinders down to the width of a small screw with no performance loss, degrading only after repeated bends around a toothpick, and a bionic-eye setup showed it caused minimal haze and no noticeable image distortion. Researchers say the platform could enable health monitors that conform to the body, curved augmented-reality displays built into windshields and pilot visors, and discreet wearables, with plans to add more complex components and improve waveguide efficiency and transparency.

• A new flagship just took the AI throne. OpenAI on Thursday began rolling out GPT-6 Astra, its newest and most powerful model, first to companies in its Daybreak cybersecurity program, with ChatGPT Plus, Pro, Business and Enterprise users, the OpenAI API, Microsoft Azure and AWS Bedrock to follow in the coming days, while free-tier users are left out. OpenAI calls Astra the world's most intelligent and aligned model, state of the art on computer use, browsing, software engineering, cybersecurity, science and professional work, and the company backed that with numbers: it saturates FrontierMath Tier 4 with a 98% score, ARC-AGI-3 with 99.9% and ExploitBench with 100%, posts 72.6% on OSWorld 2.0 for computer use, 57.9% on Terminal-Bench 4.0 for coding, 96.0% on GPQA Diamond, and perfect eight-needle long-context retrieval at 256K to 512K tokens with 96.3% out to 1 million. The model can autonomously handle tasks from website creation and scientific analysis to tax preparation and game development, cutting one sample research chore from 30 minutes of human time to five and a half. API pricing lands at $10 per million input tokens and $50 per million output, and OpenAI says Astra is its most aligned system yet, never exceeding its authorized scope in a stress test where its predecessor did so 48% of the time. President Greg Brockman closed the launch briefing by declaring, "Welcome to the AGI era."

🔊 Industry Insights & Updates

• IBM's quantum chips learned to reset in a nanosecond blink. IBM unveiled the Nighthawk r2, its fastest quantum processor yet, running over 7,500 gates and hitting 100,000 circuits per second, 25 times faster than the Heron it succeeds, without compromising scale or accuracy. The leap comes from replacing the conditional reset, which forced hundreds of microseconds of idle time between runs, with a dissipative reset gadget that uses a high-dynamic-range coupler to pull each qubit to its ground state on demand, dropping idle time to one microsecond and cutting initialization errors 25-fold. The square-lattice processor can reset qubits between circuits and mid-execution, is available now on the IBM Quantum Platform, and forms a 2026 milestone on IBM's roadmap toward billion-circuit machines by 2033.

• A "fast lane" for quantum data is open. Several Chinese research teams developed and tested a coherent quantum routing system for bucket-brigade QRAM on Origin Wukong, China's third-generation superconducting quantum computer, achieving up to 98% transmission efficiency with a single router and 93% across a two-layer network, Science and Technology Daily reported. Random-access tests separately showed average fidelity of 94.8% for one router and 82.4% for the network. The researchers say the result offers a new approach to scalable quantum memory on existing hardware, with potential to cut computing costs and improve the integration of quantum chips and circuit design.

• A stellarator startup made Japan's shortlist. METI picked Helical Fusion as a final candidate for a program backing private fusion developers, part of a three-year, roughly 60-billion-yen ($400 million) initiative running through 2028 that covers tokamak, stellarator and laser approaches, with the company's award amount still pending a formal grant decision. Helical Fusion, a 2021 spin-out of National Institute for Fusion Science research, plans to demonstrate its Helical Stellarator design through the Helix Program, progressing from superconducting magnet and blanket-divertor work to the Helix HARUKA demonstration device and eventually Helix KANATA, its first power-generating plant. More than 30 Japanese companies are already involved, and magnet demonstration work is underway in Toki, Gifu Prefecture.

• Nuclear reactors may soon run on 3D-printed parts. 3D Systems and Savannah River National Laboratory have signed a Cooperative Research and Development Agreement to develop advanced additive-manufacturing technologies for nuclear energy, critical infrastructure and national security. The partners will develop new materials, manufacturing equipment, artificial intelligence and machine-learning tools, and scalable production processes, focusing on high-temperature nickel-based superalloys and radiation-tolerant materials for components such as heat exchangers with intricate internal cooling channels, reactor internals, pumps and valves. Because 3D printing can produce geometries that are difficult or impossible to make conventionally, while cutting material waste, the partners say it could strengthen domestic manufacturing capacity and supply-chain resilience as the US invests in advanced and small modular reactors. 3D Systems equipment is already installed at the facility, and the collaboration also includes workforce training to build a pipeline of specialists for the nuclear, aerospace, defense and energy industries.

Subscribe now

🧬 BioTech

• An arthritis pill regrew hair. In two international clinical trials, the once-daily drug upadacitinib showed significant potential for people with severe alopecia areata, the autoimmune condition in which the immune system attacks hair follicles. Researchers found that after 24 weeks, roughly half of the patients taking either a 15- or 30-milligram daily dose regained at least 80% of their scalp hair, and some achieved complete or near-complete regrowth, a striking outcome given that participants had lost about 84% of their scalp hair on average. The findings were published August 12 in JAMA Dermatology, and the researchers caution that longer-term follow-up is needed to confirm how durable the regrowth is and how the benefits compare with potential risks.

• Three Seattle institutions are rewriting nature's playbook. The Allen Institute, University of Washington and Fred Hutch Cancer Center announced AI BioDesign, a collaborative accelerator that uses AI, large-scale experiments and open science to design biological solutions evolution never produced. Led scientifically by Nobel laureate David Baker and genome scientist Jay Shendure, and supported by the Paul G. Allen-backed Fund for Science and Technology, the platform will run design-build-measure-learn cycles in which AI proposes biological designs, scientists test them at scale, and results feed back to sharpen the models. All models, datasets, assays and tools will be shared openly, with goals spanning new cancer and neurodegeneration drugs, plastic-eating enzymes and ultra-efficient biological computers.

• A medical trial running for more than a decade has opened the way for potentially life-saving treatment options for patients with mild-to-moderate heart conditions. The trial examined whether an implantable cardioverter-defibrillator, a small device that detects and corrects dangerous heart rhythms that can cause sudden cardiac death, could help patients beyond the severely reduced heart function for which it is currently offered. The research focused on patients with mild-to-moderate dysfunction, a left ventricular ejection fraction of 36 to 50 percent, and heart muscle scarring identified by cardiac MRI, testing whether ICD implantation could reduce sudden cardiac death or serious ventricular arrhythmia in this group. Across all age groups the primary endpoint was not significantly reduced, but a pre-specified analysis showed benefit in patients younger than 70, and ICD implantation cut the secondary endpoint of sudden cardiac death across the full cohort.

  ---

💡Products/tools of the week

• Surfacing viral hooks and trending content formats just got easier. LightReel, an AI-powered UGC marketing research platform, continuously monitors TikTok and Instagram to help brands discover what's performing well in real time. Users can search for influencers by niche, follower count, and location, while receiving AI-generated scripts and strategy recommendations based on live social data. The platform also analyzes uploaded videos and brand accounts to deliver instant feedback. A developer API lets teams automate research tasks like hook discovery, creator sourcing, and competitor analysis at scale.

• Crowdsourced AI video entertainment is now streaming live. Infinite Slop, is a continuous AI-generated video channel where viewers collectively decide what content airs next by submitting and upvoting prompts in a live chat. The platform uses generative AI to produce short video clips based on winning chat suggestions in real time. The result is an unpredictable, community-driven stream of absurd, humorous, and surreal AI-generated scenes. Hundreds of viewers watch simultaneously, making it a social entertainment experiment powered entirely by generative video AI.

• Clicking to edit text inside images no longer requires design skills. PhoText, an AI-powered online editor, lets users directly modify or replace text seamlessly. The platform uses smart background repair and style transfer to match fonts and shadows perfectly. It includes high-precision OCR for all languages and conversational editing across multiple file formats on web and mobile. Basic features are free with a generous trial quota for advanced capabilities.

• Querying multiple academic databases at once is possible through a single API. Minicod, a unified academic data API platform, aggregates Semantic Scholar, OpenAlex, and PubMed through one consistent JSON response schema with DOI and open-access links. The service offers an OpenAI-compatible chat completions endpoint and an MCP server for Claude Desktop, Cursor, and other AI agents. It also includes citation graph traversal, journal metadata, and a Skills Marketplace of reusable research-workflow instruction bundles.

  ---

• zeroth_official: Zeroth Bridge Reveal: The $4,000 High-Performance Personal Humanoid Robot

• ROBOTIS OpenSourceTeam: AI Sapiens #6: OH! GYM! Students Bring Humanoid Motions from Simulation to Reality

• Galileo Robotics: Galileo X | Defy Form Limits, Transcend Imagination

Welcome to this week's edition of Rushing Robotics! If you blinked this week, you missed at least three world-firsts. A full-stack MRI brain-computer interface that reads and writes the brain. The first AI-assisted live brain surgery, which ended with a man keeping his sight. A one-dose gene therapy for inherited heart failure. Also: OpenAI's chip outmuscled Nvidia, IBM taught a mainframe Arm, and humanoid robots took up tennis. Here's everything that mattered.

🤯 Mind-Blowing

From a machine that reads and writes the brain in one closed loop, to an AI second pair of eyes watching over a neurosurgeon's shoulder mid-operation — these are the stories that made us do a double-take this week. Add a gene therapy that restores broken hearts at the cellular level, a custom chip that beats the industry's best on performance per watt, and robots that now come with their own automated training coach, and you've got a week the future won't forget.

🔊 Industry Insights & Updates

The plumbing of the future got serious upgrades this week. IBM's 2-nanometer dual-architecture processor brings Arm's 22-million-developer ecosystem to the mainframe for the first time, a whisper-thin layer of salt pushed perovskite tandem solar cells past 30% efficiency, a nine-seat hybrid aircraft that lands on a soccer field is heading into production, and New York now hosts America's longest wireless quantum link — 13 miles of entanglement-ready light through open air.

🧬 BioTech

Healing is becoming programmable. This week's lineup: a nanoparticle hydrogel that nearly sealed a chronic wound in 72 hours, engineered tRNA that reads through faulty stop signals and could one day treat thousands of genetic diseases with a single strategy, and a pill doctors already prescribe that may halt the spread of one of the deadliest breast cancers.

💡 Products/Tools of the Week

Your workflow just leveled up. Build a cinematic 3D website from a single sentence, plug AI agent harnesses into your product with one API call, get a clear ship-or-block security verdict on AI-built apps, and hand everyday busywork across 1,500+ apps to plain-English AI agents.

🎥 Video Section

Grab the popcorn — the 2026 World Humanoid Robot Games delivered everything this week: an intense 1-on-1 humanoid boxing match, the world's first fully autonomous robot tennis rally, and a glorious highlight reel of fails proving the robot revolution still has a few bugs left to iron out.

Weeks like this are a reminder that the future rarely arrives with fanfare — it compounds quietly, one brain interface, one gene therapy, and one improbable robot tennis match at a time. The tools being built today are collapsing the distance between what we imagine and what we can actually do, and that pace is only accelerating. Stay hungry, stay futurish!

🤯 Mind-Blowing

• Reading the brain and writing to it now happen in one machine. China has released uMR Shenguan, described as the world's first full-stack MRI-based brain-computer interface, combining signal acquisition, decoding, neuromodulation, and evaluation in a single closed-loop system. UIH and TJU jointly built the platform, in which the MRI scanner serves as the loop's perception and feedback hub, capturing neural signals at millisecond speed and brain structures at sub-millimeter precision while compensating for interference from implanted devices. The developers say the system makes neural plasticity measurable and controllable, with clinical uses emerging in rehabilitation and psychiatry and consumer applications expected to follow.

• Brain surgery has a new second pair of eyes. For the first time, real-time AI assisted a neurosurgeon during a live operation, helping remove a pituitary tumor while sparing the patient's vision. UCL and NHNN surgeons ran the clinical-trial procedure in London using UCL-built software that analyzed the endoscopic video feed as it happened, flagging risky anatomy in a space where a millimeter's error can cause blindness or stroke. Hani Marcus led the operation with Danyal Khan, the surgeon fully in control throughout. Patient Rhys Hibbert, 48, whose tumor had stolen much of his peripheral sight, walked without glasses or sticks within a week. Backed by the NIHR and Google, the UCL team next wants the system to track instruments and monitor tissue interaction in real time.

• Genetic heart disease has met its first real challenger. A single-dose gene therapy restored normal heartbeat strength and rhythm in patient-derived heart tissues and reversed disease in mice, according to work published in Nature Cardiovascular Research. MCRI scientists replaced faulty copies of the ALPK3 gene with healthy ones, fixing the internal scaffolding of heart cells; AI-assisted analysis suggests the same strategy could help patients with MYH7 and TTN-related cardiomyopathies, among the most common genetic causes of heart failure. With about 30 million people affected by cardiomyopathy worldwide, the team is preparing the safety studies needed before human trials and seeking commercial partners to get there.

• A custom chip just outmuscled the industry's best on efficiency. OpenAI's first in-house inference chip, Jalapeño, delivered 1.5 to 1.9 times more AI work per watt and 1.7 to 3.6 times lower latency than Nvidia's top commercial systems in new benchmark tests. OpenAI measured the chip on the public InferenceX benchmark from SemiAnalysis across three large models — its own GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2.5 1T — and reported 2.1 to 4.1 times higher performance on highly interactive workloads. Rated at 700 watts but sustaining 550 watts or less in testing, Jalapeño is built to erase the usual tradeoff between throughput and latency: it handles both the compute-heavy prefill phase and the bandwidth-hungry decode phase in one architecture, keeping model state close to its processors and integrating networking to cut data movement between chips. OpenAI designed it with AI's help, moving from initial design to tapeout in nine months, and plans deployment in its compute infrastructure by the end of 2026 as the first of a multigenerational roadmap.

• Robots learning to navigate new places now have an automated coach. A new framework called COMPASS reduces the rework normally required when robots, scenes, or conditions change. NVIDIA researchers paired the company’s pretrained X-Mobility policy with residual reinforcement learning, training a specialist that adjusts the base policy for each robot-environment pair rather than starting over. AI coding agents run the pipeline, preparing simulation assets, testing, training, debugging failures, and comparing checkpoints, while humans approve the critical decisions. NVIDIA validated the workflow on the Spot quadruped, measuring goal success, falls, and route time, and says approved policies can generate movement commands directly from camera images, odometry, and a target goal.

🔊 Industry Insights & Updates

• Arm is moving into the mainframe. A new IBM processor built on a 2-nanometer process can run Arm-native Linux applications alongside traditional z/OS and IBM workloads on the very same cores, a first for enterprise computing. IBM and Arm gave the chip 11 cores running above 5.7 GHz, each able to execute both instruction sets natively, plus AI inference accelerators for real-time fraud detection and a dedicated on-chip unit for I/O acceleration. The design, the first hardware result of the companies' April collaboration, targets future IBM Z and LinuxONE systems and opens the door to Arm's 22-million-developer ecosystem.

• A tiny salt layer just fixed a big solar problem. Cesium chloride, deposited as an ultra-thin seed layer, has pushed perovskite-silicon tandem solar cells to 30.3 percent efficiency while making them far easier to mass-produce. HZB scientists had faced years of manufacturing frustration: on textured silicon, the self-assembling monolayer molecules that transport positive charges collected in the surface valleys and left the peaks exposed, and perovskite grown on top developed defects that sapped power. Nanoscale imaging at the BESSY II synchrotron revealed the failure, and the seed layer corrected it by leveling the microscopic terrain and guiding uniform crystal growth, as lead author Viktor Škorjanc explains.

• Regional air travel may soon skip the airport entirely. A nine-passenger hybrid-electric aircraft able to take off and land on a soccer field is heading into production in Springfield, Ohio, through an $850 million investment creating nearly 2,000 jobs. MIT-born startup Electra pairs a nose-mounted gas generator with under-floor batteries and eight electric wing motors, using blown lift for ultra-short takeoffs and quiet operation in noise-sensitive areas. The plane flies 1,200 miles at around 200 mph, targeting the 50-to-250-mile trips where cars and airlines both fall short. Electra's EL2 demonstrator has logged over 200 test flights since 2023; the new factory breaks ground next year, aiming first at 400 aircraft a year, then 800.

• America's longest quantum network just went wireless. Quantum states of light — and later entangled photons — traveled 13 miles through open air between rooftops in Stony Brook and Upton, New York, a US first. Brookhaven Lab (BNL) and Stony Brook University (SBU) researchers built the link around a laser source at SBU's Quantum Watchtower, an ultrafast camera at BNL's Quantum Lighthouse, and adaptive optics from astronomy that kept atmospheric turbulence from destroying the signal. The free-space optical connection won't replace fiber; it extends the existing 161-mile, eight-node network to places cables can't reach and to wavelengths fiber doesn't favor. A Yale facility comes next, enabling a 30-mile hop across Long Island Sound, with satellite links as the long-term goal.

Subscribe now

🧬 BioTech

• A tiny particle is taking on wounds that won't close. In lab tests, a hydrogel loaded with hybrid nanoparticles nearly sealed a simulated chronic wound in just 72 hours. USP researchers designed the particles to deliver two agents at once: siRNA that switches off production of MMP-9, an enzyme that eats away at skin-repair proteins in chronic wounds, and resveratrol, which calms inflammation and spurs collagen-making fibroblasts. The nanoparticle protects both fragile molecules and releases them slowly at the wound site. The FAPESP-funded team says patients with diabetic and pressure ulcers, who face heightened risks of infection and amputation, could eventually benefit, though 3D skin models and animal studies must come first.

• One drug strategy could soon serve thousands of genetic diseases. Engineered transfer RNA has read through the premature stop signals known as nonsense mutations, restoring functional protein in cystic fibrosis models where today's drugs fail. U of T researchers paired chemically enhanced tRNA, made longer-lasting with a modification borrowed from nature, with a tailor-made lipid nanoparticle delivery system, publishing the results in Science. The restored CFTR protein kept working for over 40 days in human airway cells, and in patient-derived organoids the tRNA combined with Trikafta succeeded where neither worked alone. Study lead Bowen Li says the goal is tRNA medicines that recognize shared stop signals across many diseases, with an inhalable version already in sight.

• A breast cancer killer may be stopped with a pill doctors already prescribe. Restoring levels of a natural molecule called miR-342 sharply slowed the spread of triple-negative breast cancer to the lungs and bones in pre-clinical models, and the existing CDK4/6 inhibitor palbociclib achieved similar results where miR-342 was low. AU and ONJCRI researchers identified the miR-342–E2F network as the switch controlling metastatic growth, reporting in EMBO Molecular Medicine that low miR-342 combined with high E2F activity flags the patients most at risk. That same signature could serve as a test to select triple-negative patients for CDK4/6 inhibitor treatment, repurposing an existing therapy for one of the deadliest breast cancers. Validation in patient-derived models comes next, before clinical trials.

  ---

💡Products/tools of the week

• A full cinematic 3D website now takes a single sentence. Draftly, an AI-powered 3D website builder, turns one text prompt into a scroll-driven web experience, requiring no coding or design skills from the user. The platform handles the entire pipeline on its own: it generates visuals, animates motion, extracts video frames, and assembles everything into a scroll-reactive 3D experience, then exports production-ready HTML, CSS, and JavaScript. Users who want changes can chat with the AI to adjust copy, colors, and layout live, or switch to a node-based visual studio for deeper customization of the build. Finished projects publish directly to a Draftly subdomain or a custom domain, or download as a ZIP file for self-hosting. A free plan is available, with paid plans starting at $25 per month.

• Agent-powered features just became an API call away. HarnessRouter's unified platform lets developers plug AI agent harnesses such as Codex, Claude Code, and Hermes into their products, running each session in an isolated serverless sandbox while it manages provisioning, streaming, retries, and artifact delivery. A benchmarking tool compares harness and model combinations on cost, latency, and success rate, and one API key is all it takes to start shipping.

• Shipping an AI-built app now comes with a clear go or no-go signal. AtRisk, a continuous security tool, scans live URLs and connected GitHub repositories for vulnerabilities across 100+ checks, then returns a Ship/Block score indicating whether an app is safe to deploy. Findings from the live site and source code land in a single inbox, paired with paste-ready fix prompts that work with Cursor and Claude Code via MCP. Automatic pull request reviews come built in, and the Pro plan can block risky deployments via a GitHub Action with SARIF output.

• Everyday work across an entire app stack can now be handed to AI agents. DeskFerry, an AI automation platform, turns plain-English instructions into executed tasks across more than 1,500 connected apps, offering DeskFerry Assist for quick, repeatable jobs and an Agent Builder for multi-step workflows spanning several tools. Connections include Gmail, Slack, HubSpot, Notion, and Salesforce, with AI models from OpenAI, Anthropic, Google, and others available without separate API keys. Custom skills keep agents on-brand, format outputs, and file results into the right tools automatically.

  ---

• New York Post: Humanoid Robots Battle in Intense 1-on-1 Fight in China

• Galbot: From AlphaGo to AstraTennis: The World’s First Autonomous Humanoid Robot Tennis Match | GALBOT

• news.com.au: Biggest fails from the 2026 World Humanoid Robot Games in China

Welcome to this week's edition of Rushing Robotics! If this week proved anything, it's that the line between science fiction and an ordinary Friday is getting dangerously thin. Scientists just collected $1 million for reversing biological aging in mice, a teenager's metastatic cancer vanished after a single infusion of reprogrammed immune cells, robots learned to "feel" objects without ever touching them — and a home humanoid went on sale for less than a MacBook. Meanwhile, bacteria are being wired into living circuits, an entire neighborhood is rising from robotic printers in Colorado, and IBM is chaining together ultra-cold modules on the road to fault-tolerant quantum computing. Grab your coffee — this week's issue is a ride.

🤯 Mind-Blowing

The stories that made us stop scrolling and stare. This week: super-exosomes that hunt down aging cells win a $1M XPRIZE milestone, a single T-cell infusion wipes out metastatic kidney cancer, eel-inspired sensors give robots a sixth sense, a robot brain survives a full body swap, and a $1,688 humanoid promises to fold your laundry.

🔊 Industry Insights & Updates

Where the money, machines, and megawatts are moving. MIT prints living transistors from bacteria, robots break ground on a 55-acre printed community, IBM fuses cryogenic modules toward a 1,000-qubit system, and Nvidia's $105 billion guarantee for OpenAI's Ohio megasite redraws the map of the AI infrastructure arms race.

🧬 BioTech

Medicine's quiet revolution marches on: human heart valve tissue grown from stem cells in a world first, vision restored in patients with a previously untreatable genetic eye condition, and an unexpected bacteria–immune system alliance that could open an entirely new front against cancer.

💡 Products/Tools of the Week

Fresh gear for builders and operators: turn text prompts into production-ready 3D models, track exactly what AI assistants say about your brand, surface high-intent leads hiding in public communities, and run voice AI entirely on your own infrastructure.

🎥 Video Section

This week's must-watch footage comes straight from China's humanoid frontier — superhuman abilities, a major Terminator-style upgrade, and a more dexterous, more practical TienKung 3.0. Seeing is believing.

We say it often, but it bears repeating: we are living through the most extraordinary chapter in human history. Cancers that were once death sentences are becoming curable, aging itself is being treated as an engineering problem, and machines are learning to see, feel, heal, and build right alongside us. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A new contender has emerged in the global push to turn back the biological clock. Goda Lab, has been selected as one of the Milestone 2 Awardees in the XPRIZE Healthspan competition, collecting US$1 million to fund further testing after being chosen from more than 600 teams worldwide. The lab's approach centers on super exosomes: natural cell-to-cell transport particles whose surfaces are chemically engineered with lanthanide ions so they can efficiently seek out senescent cells — old or damaged cells that linger in the body and fuel aging — and then unload regenerative cargo such as growth factors or genetic material directly into aging tissue. Across multiple cohorts of aged mice, the therapy significantly improved frailty, physical performance, and other aging-associated measures while maintaining an excellent safety profile, with cross-species scaling suggesting a benefit of more than 15 healthy human years. The team will now begin clinical studies aiming to prove the technology can extend healthy human lifespan by at least a decade — the bar for the XPRIZE Grand Prize. Beyond healthy aging, the platform could find uses in preserving cognitive, muscular, and immune function as well as hair regeneration, skin rejuvenation, and reproductive medicine, with commercialization planned through NanoTitan to establish scalable manufacturing and global distribution of the Japanese-developed therapeutics.

• A complete remission of metastatic kidney cancer has been achieved in a teenager with a single infusion of reprogrammed immune cells. The team at KiTZ Heidelberg genetically modified the patient's own T cells to target the PRAME antigen, with the reprogramming vector supplied by collaboration partner Immatics and the cells manufactured in just seven days at NCT Heidelberg. Tumor regression followed across all affected organs, and three months after treatment no living tumor cells remained. "At the present time, almost one year after the infusion of the T cells, our patient is in excellent condition. He trains regularly, takes part in cycling races, was able to successfully complete his vocational training, and now plans to obtain his Abitur," said treating physician Christian Seitz of KiTZ. The case, published in the New England Journal of Medicine, underscores the potential of cellular immunotherapies for children with solid tumors.

• Robots can now "feel" objects without ever touching them. Chinese researchers have developed a new sensor inspired by electric eels that uses weak electric fields, distorted by nearby objects, to enable touchless detection and identification. Just as an eel generates an electric field in water and reads the changes caused by fish, rocks, or plants, the sensor built by Professor Zhang Weiqiang's team at Xidian University uses a charged fluoropolymer surface — essentially a tiny static battery that holds its charge for long periods — to create a monitorable electric field. When an object approaches, the field's shape warps in measurable ways, letting the system infer the object's electrical conductivity, dielectric properties, geometric shape, and even its proximity. "We want the machine to sense an approaching target – distinguish its material and surface condition – before any physical contact," Zhang said. The technology could prove valuable for delicate manipulation, prosthetics, and manufacturing robots, especially in darkness, smoke, dust, or other settings where cameras fall short.

• A robot brain that survives a body swap has been unveiled. Feagine Robotics has introduced Fi0, a cross-embodiment foundation model designed to retain task knowledge across robots with different physical structures, alongside three tendon-driven soft manipulators — A01, A02, and A03 — built to test the idea. Rather than treating each machine as a separate system, Fi0 feeds the robot’s physical structure and current state directly into the information used to generate actions, so the task stays the same even when the body accomplishing it changes. The company says a single human demonstration can teach Fi0 a new task at inference time, with no retraining cycle: the model extracts the task’s objects, sequence, contact events, and desired end state, then works out how to achieve it with whatever body it inhabits. Feagine calls the approach soft embodied intelligence, and while Fi0 remains an early-generation system, it targets a genuine bottleneck in robotics — ensuring intelligence learned on one machine doesn’t become useless when the hardware changes.

• A humanoid robot for household chores has hit the market at just $1,688. San Francisco-based Nori Robotics launched the American-assembled Nori A3 this week, with shipping set to begin this fall for consumers seeking affordable home automation. Unlike many humanoids, the A3 moves on a wheeled base and features 19 degrees of freedom, with a central body that raises and lowers its head and arms. Nori says the robot can fetch food from a refrigerator, pour drinks, load a dishwasher, and fold clothes, lifting up to 122 pounds vertically, though payload drops to 3.3 pounds with arms fully extended. Lidar and four 720p cameras positioned around its hands, head, and neck handle perception and navigation, while a speaker and microphone enable spoken commands. The battery delivers six to eight hours of operation, and a Skill Marketplace lets owners share and install capabilities developed by other users.

🔊 Industry Insights & Updates

• A new kind of circuit board has come to life — literally. MIT engineers have created biological transistors from living bacteria and printed them into functioning logic circuits on agar, with chemical molecules flowing between colonies in place of electrons through wires. The modular design sidesteps a chronic problem in synthetic biology: instead of cramming many functions into one cell and overloading its protein-making machinery, the work spreads computation across separate cells that each handle a simple task. The approach, published in Nature Chemical Biology and funded in part by DARPA and IARPA, produced circuits capable of multi-input logic, addition, and signal routing — all from just five bacterial strains whose arrangement, not their genetics, determines the computation.

• An entire neighborhood is being built by robots in Colorado. Cleora, a 55-acre development in Salida, is taking shape as one of the world's largest planned 3D-printed residential communities, with more than 65 robotically constructed homes and infrastructure managed through its own metropolitan district. Developed by Cleora with 3D-printing contractor True North and technology provider RIC Robotics, the project follows nearly seven years of research into construction technology and scalable housing, and aims to prove robotic construction can move beyond prototypes to support real communities. Two autonomous RIC printing systems are already operating across the site, and True North's first model home is under construction with its shell expected to be completed this month. "From day one, our vision wasn't simply to build 3D-printed homes, it was to prove that robotics could transform how entire communities are designed and delivered," said Greg Kenny, Managing Partner at Cleora. The development will also serve as a training hub through a partnership with Colorado Mountain College, preparing workers for an increasingly automated construction industry.

• Two cryogenic modules have been successfully joined and cooled into a single ultra-cold environment, IBM announced, in a demonstration that brings fault-tolerant quantum computing a step closer to reality. The box-shaped modules reached 4 Kelvin in under five days and a final temperature below 15 milliKelvin shortly after, while offering 12 times more wiring space than IBM's earlier systems. Built to house hundreds of quantum chips, the modular design separates components and subsystems so each can be independently optimized as wiring and device density grow. IBM plans to begin operational testing with its NightHawk quantum processors installed in the modules later this year, supporting its goal of a 1,000-programmable-qubit system by 2027.

• The AI infrastructure arms race has a new front line. OpenAI has locked in a 20-year lease on a massive Ohio data center developed by SB Energy, with Nvidia guaranteeing up to $105 billion in lease and power obligations — a structure critics quickly flagged as "circular" financing in which suppliers, buyers, and investors prop up one another's balance sheets. Nvidia CEO Jensen Huang pushed back firmly: "Is this circular financing? No," he declared, stressing that OpenAI remains directly on the hook for rent payments while Nvidia simply uses its scale to secure durable sites that can be upgraded across multiple chip generations. "AI is becoming infrastructure – the foundation for intelligence in every industry," Huang said. The Pike County project, launching at 4.25 gigawatts in 2028, follows last year's $500 billion Stargate announcement and surpasses Meta's planned 5-gigawatt Louisiana facility and SpaceX's 1-gigawatt Colossus network.

Subscribe now

🧬 BioTech

• Human heart valve tissue has been grown from stem cells in a world-first breakthrough. The discovery, led by the Murdoch Children's Research Institute (MCRI) in Melbourne and published in Cell Stem Cell, produced lab-grown valve tissues that closely resemble those in the human body, advancing understanding of how heart valves develop, mature, and become damaged. Heart valve disease affects about 28 million people worldwide, and with no treatments able to reverse the damage, most patients depend on replacement surgery that carries a high risk of blood clots and, for children, the likelihood of repeated procedures over a lifetime. "By creating heart valve tissues that mimic the molecular features of a human version we can now test and advance new treatments at scale in the lab," said MCRI's Dr Holly Voges. The team also showed the platform can model inflammatory valve disease, a precursor to rheumatic heart disease, with engineered tissues becoming stiffer and showing disease biomarkers when exposed to inflammatory proteins.

• Vision has been restored in patients with a previously untreatable rare genetic eye condition. A first-in-human clinical trial at Moorfields Eye Hospital and University College London (UCL), published in JAMA Ophthalmology, tested a breakthrough stem cell treatment for aniridia-related keratopathy (ARK), a progressive disease caused by a PAX6 gene mutation that affects one in 40,000 to 100,000 people and gradually turns the cornea opaque. The Moorfields and UCL team developed a collagen scaffold tissue-engineered to deliver limbal epithelial stem cells and stromal stem cells to the affected area, treating one eye in each of nine patients while the other served as a control. Treated eyes improved from an average ocular surface score of 9.4 to 5.9 after three months, while untreated eyes barely changed, and patients gained an average of 24 letters — almost five lines — on a standard vision chart. The team now plans a larger trial with a view to NHS adoption.

• An unexpected alliance between bacteria and the immune system has been uncovered in lung tumors. Johns Hopkins researchers discovered that adding a vitamin B2 metabolite to tumor-dwelling bacteria sharply increases MR1 protein expression on immune cells, switching on MAIT cells — innate immune cells that have received far less attention than the adaptive T cells targeted by today's checkpoint inhibitors. Surprisingly, even enterococci bacteria that don't produce the B2 metabolite themselves greatly enhanced MAIT activation once the metabolite was added. "Tumor-associated bacteria may regulate the immune response to tumors by regulating the expression of MR1 on antigen-presenting cells and activating the MAIT cells in the tumor microenvironment, which could have anti-tumor effects," explained Pakhi Birla, who led the study. The findings, published in PNAS, point toward a new way to stimulate the immune system to attack cancer.

  ---

💡Products/tools of the week

• Text prompts and plain images can now be transformed into production-ready 3D models inside a single browser-based workspace. Trify3D, a multi-model AI platform, makes this possible by running leading generation engines such as Tripo3D, Meshy, and Rodin side by side, so users can compare mesh outputs before committing any credits. Once a result is chosen, it exports in GLB, glTF, OBJ, or STL formats, ready for Unity, Unreal, Blender, or 3D printing slicers. Trify3D also includes AI texturing with full PBR map output, covering albedo, normal, roughness, and metallic channels. A single shared credit pool spans every supported engine, and no software installation is required.

• A new level of transparency has arrived for companies trying to understand how AI talks about them. Lettertrace has released an open-source, bring-your-own-key tool that tracks how often AI assistants such as Claude, ChatGPT, and Gemini mention a brand across auto-generated prompt variations. Rather than relying on spot checks, users get structured metrics on visibility, share of voice, prominence, and sentiment, broken down per topic, per model, and over time. Lettertrace supports competitor benchmarking and scheduled automated monitoring runs, and because the whole stack self-hosts on the user's own Supabase instance, no data ever leaves their control. The MIT-licensed CLI and web app are built for AEO and GEO teams who want full data ownership and zero vendor lock-in.

• High-intent leads hiding in public communities can be surfaced through a single unified workflow. Needle, a buyer-intent search platform, scans Reddit, Hacker News, Stack Overflow, GitHub, X, and LinkedIn, ranking results by buying intent rather than raw keyword matches and filtering out irrelevant noise automatically. Users can run manual one-off searches or schedule automated monitoring to track brand mentions and competitor signals. Needle integrates with CRMs like HubSpot and Pipedrive and delivers notifications via Slack, Discord, and Google Chat, making it a practical fit for founders, sales teams, and marketers doing community-led customer discovery.

• Voice AI no longer has to mean vendor lock-in or data leaving your servers. Dograh, an open-source, self-hostable voice agent platform, lets teams build and deploy AI-powered voice agents with their own STT, LLM, TTS, and telephony providers, including fully on-premise setups, as an alternative to hosted services like Vapi and Retell. Dograh supports cascaded pipelines and direct speech-to-speech models across 70+ languages, ships an MCP server for configuring agents from tools like Claude Code or Cursor, and keeps HIPAA, GDPR, SOC 2, and data sovereignty compliance within the user's own infrastructure.

  ---

• AI Revolution: China Just Dropped Superman - AI Robot With Superhuman Abilities

• NextGen Humanoids: China’s Terminator Robot T800 Just Got a MAJOR Upgrade… (EngineAI New Tech)

• TienKung Robotics: Say Hi to TienKung 3.0 ：More open / More Dexterous / More Practical 🙌🙌🙌🤖🤖🤖

Welcome to this week's edition of Rushing Robotics! This week, a robotic hand out-twisted and out-reached its human counterpart and scientists reconstructed a complete human genome from both parents. Add in faster AI agents, quantum-proof chips, and a hydrogen engine with diesel muscle — the future isn't waiting.

🤯 Mind-Blowing

Some weeks redefine what machines can do — this is one of them. China's BioflexBot skips biological fingers entirely, using a coiled spring, a shell, and compressed air to thread acupuncture needles, run chemistry experiments, and reach 3.5 times farther than a human hand. ARPA-H is funding fully autonomous stroke-intervention robots to fix a stark gap: only 12% of acute ischemic stroke patients receive gold-standard thrombectomy. Plus: DYNA-2 learns physical intuition from a million hours of human video, a post-quantum chip passes validation in silicon, and hydrogen combustion gets serious.

🔊 Industry Insights & Updates

The business of the future kept pace. Agility Robotics ships its Digit V5 humanoid to early customers in December, promising 20-hour workdays without safety cages. NVIDIA's open Nemotron 3.5 Lightning activates just 3 billion of its 30 billion parameters per token, making autonomous agents faster and cheaper. Researchers pinned silicon qubit variability on atomic-scale disorder — a solvable manufacturing problem — while OpenLight and Tower Semiconductor expanded their platform to accelerate 1.6T photonic chips for AI data centers.

🧬 BioTech

Biology delivered its own share of breakthroughs this week. Scientists reconstructed a complete diploid human genome — telomere to telomere, with both parents' chromosome copies intact — revealing 15% more of the genome than prior benchmarks, including previously inaccessible regions relevant to cancer and neurological disorders. Cancer researchers gained access to a compendium of 665 cell line and organoid models paired with their original human tumors, rigorously analyzed to ensure close fidelity to their source cells. And in transplant medicine, University of Pennsylvania researchers used an engineered cold virus carrying the CTLA4Ig gene to trick the immune system into permanently accepting a donor organ — by blocking the costimulatory signal T cells need to attack foreign tissue. Personalized, precise, and increasingly practical: medicine's next era is arriving.

💡 Products/Tools of the Week

Unsloth Desktop became the first desktop app to both run and train AI models locally — no Python environments, no terminal commands, with training at 2× speed using 70% less VRAM. Cactal gives AI agents a single API to build, host, and operate websites at scale, with a nine-point quality control pipeline before anything publishes. Tracetify reverse-engineers exactly how competitor websites grew from scratch, drawing on 12 data sources and over 100 linked references. And FeedLog offers an open-source, AI-native alternative to Canny, with semantic duplicate detection and an AI changelog assistant that writes release notes in your brand voice. Builders, take note.

🎥 Video Section

Prefer to watch the future unfold? This week's picks have you covered: CNET introduces a flying, swimming robot inspired by aquatic birds, NVIDIA shows how Isaac GR00T is accelerating humanoid robot development, and Fireship spends three days at MIT arguing the robot hype is worse than you think. Three perspectives, one theme — embodied AI is having its moment.

Robots that learn from watching us, genomes read end-to-end, and chips built to outlast quantum threats are all pieces of the same accelerating puzzle. The gap between laboratory breakthrough and real-world deployment has never been narrower, and each week it closes a little more. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A robotic hand that threads acupuncture needles, operates a pipette, and twists bottle caps nearly four times farther than a human hand has been revealed in China. Researchers led by senior author Yingtian Li, PhD, of the Chinese University of Hong Kong in Shenzhen, built the prototype called BioflexBot, taking a function-first approach by using a coiled spring, a constraining shell, and compressed air instead of reproducing biological fingers, joints, and tendons. With only two pneumatic inputs, it can pinch, rotate, hook, and grasp, extending and contracting 3.5 times more than a human hand to reach distant objects and work in narrow spaces. The team deployed it to inspect aeroengine blades, perform everyday tasks on a humanoid platform, and run a delicate chemistry experiment.

• Autonomous robots for life-saving stroke surgery got a major federal boost, as the Advanced Research Projects Agency for Health (ARPA-H) announced the teams receiving contract awards from its Autonomous Interventions and Robotics (AIR) program, with a commitment of up to $175.3 million over five years. The program responds to the fact that only 12% of the 700,000 Americans who suffer acute ischemic strokes each year receive mechanical thrombectomy, the gold-standard treatment, largely due to geographic barriers and a shortage of trained specialists. Teams must demonstrate autonomous capabilities in benchtop or biological models within 24 months and fully autonomous surgical interventions in realistic models, animals, or human cadavers within 60 months.

• Robots learning physical intuition straight from human footage is the idea behind a new foundation model, with Dyna Robotics unveiling DYNA-2, trained on more than 1 million hours of human video. The Redwood City, California-based company said its World-Action Model uses a world-modeling architecture that combines next-frame and next-action prediction, so instead of learning only from actions performed by robots, the system uses human video to develop an understanding of how physical environments change and how objects respond to movement. That lets knowledge gained from human behavior transfer across different robot hardware, according to the company, which noted that DYNA-2 had not seen robot data during pre-training yet could later be adapted to different platforms with only a few hours of local fine-tuning. Dyna Robotics also said its video co-training method improved scores by 133 percent on instruction-following tasks that required robots to perform different physical motions based on commands.

• Hardware that keeps secrets safe even when quantum computers arrive has passed a key test, with BTQ Technologies and the Industrial Technology Research Institute (ITRI) completing the first milestone in validating a new chip architecture for post-quantum cryptography. The Quantum Compute-in-Memory (QCIM) architecture is designed as a soft IP cryptographic accelerator capable of supporting both conventional and post-quantum cryptographic functions in a compact, low-power design, and is also 'crypto-agile,' meaning it can support different cryptographic algorithms as requirements change. In the first phase of the collaboration, the architecture was validated under demanding operating conditions in a TSMC 28-nanometre environment, demonstrating acceleration of cryptographic operations associated with FIPS 203, 204 and 205. The technology is being targeted at applications where long-term security, device authentication and efficient cryptographic processing matter most, including edge computing, industrial systems, secure elements and AI-enabled devices, and the program will now move into module-level integration, verification and validation.

• Medium-duty commercial trucks could soon run on hydrogen with the muscle of diesel, as researchers from the Southwest Research Institute (SwRI) developed an optimized spark-ignited hydrogen engine suitable for medium-duty commercial vehicles. The new engine is not a fuel-cell design but a true internal combustion engine that burns hydrogen inside cylinders using spark ignition, behaving much more like a diesel engine than earlier hydrogen engines. On emissions, the engine produces near-zero tailpipe CO, though the team notes this depends on how the hydrogen fuel is produced, and because hydrogen combustion can still produce nitrogen oxides at high temperatures, the engine is not literally zero-emission and may require careful combustion control and exhaust after-treatment.

🔊 Industry Insights & Updates

• Twenty-hour workdays on the factory floor are the promise of a new humanoid robot, as Agility Robotics readies its Digit V5 for deployment to early customers in December. The US-based company's machine is designed for warehouse and factory operations, where it can work alongside people without requiring the extensive safety fencing typically used around industrial robots, addressing a key barrier to wider commercial use of humanoids. Its battery system is designed to support more than 20 hours of work per day, potentially allowing one robot to operate across three shifts, and its ISO-standard flange lets it work with a broad range of robotic end-effectors. Agility says Digit V5 incorporates upgraded safety systems intended to let the robot operate in close proximity to people, removing cages and barriers that can consume valuable floor space, require costly facility modifications, disrupt traffic patterns, and force changes to existing workflows.

• A new open AI model built to make autonomous agents faster and cheaper to run has arrived, as NVIDIA launched Nemotron 3.5 Lightning, a model with 30 billion total parameters that activates only 3 billion per token thanks to a mixture-of-experts design. The model targets the execution layer of autonomous agents, where systems repeatedly call tools, check results, run commands, and delegate tasks, letting developers reserve larger reasoning models for planning and complex decisions. NVIDIA designed Lightning to run locally on DGX Spark, Jetson systems, and GeForce RTX 5090 graphics cards, as well as in data centers, and trained it with popular agent harnesses like OpenClaw and Hermes Agent in mind. The company claims up to four times the output speed of similar-sized models, and on PinchBench the model reached 86 percent accuracy while completing 10,000 tasks 30% faster than Qwen3.6 35B at similar accuracy.

• Atomic-scale disorder in silicon quantum wells has been identified as the main source of valley splitting variability, according to new research that recasts a fundamental quantum computing challenge as a solvable materials problem. The team conducted their investigation using an Intel-fabricated 12-qubit-class silicon quantum dot processor at Argonne National Laboratory’s Chicago Quantum Computing Testbed, where they utilized electrical spectroscopy to measure valley splitting while shifting quantum dots along the quantum well. By analyzing how these variations correlate with distance, the researchers pinpointed random atomic-scale fluctuations in the alloyed quantum well as the dominant factor, providing a nanoscale map that could guide manufacturing improvements for more reliable silicon spin qubits with enhanced fidelity.

• Faster development of 1.6T photonic chips for data centers is the goal of a newly expanded smart platform, with OpenLight and Tower Semiconductor announcing an upgrade to their PH18DA photonics ecosystem through the unveiling of OpenLight's photonic Process Design Kit. The California-based semiconductor company and the Israeli chipmaker say the technology can be utilized for 400G and 1.6T photonic circuits, simplifying and accelerating the creation of new photonic chips for high-speed optical communications and AI data center infrastructure. With the platform, the companies are aiming for 1.6T connections, targeting laser-integrated photonic circuits that could support advanced near-packaged optics and co-packaged optics, two approaches seen as critical for next-generation data center performance.

Subscribe now

🧬 BioTech

• Personalized genomics took a leap toward everyday medicine, with scientists reconstructing the complete genome of a real person, with full sets of chromosomes from each parent. The findings build on the groundbreaking 2022 completion of the first complete human genome by the T2T Consortium, but the new studies go much further by reconstructing complete "diploid" genomes, meaning each genome contains two slightly different copies of every chromosome, with one copy inherited from each parent. The team sequenced the HG002 human genome, a sample from a living donor widely used as a reference material by the DNA sequencing and diagnostics industries, and with near-perfect accuracy, each chromosome spans from "telomere to telomere" and reveals 15% more of the genome, including previously inaccessible portions relevant to cancer and neurological disorders. The team added more than 900 million DNA letters that were absent from the prior benchmarks, including both sex chromosomes and areas with genes known to affect disease risk.

• A trove of 665 tumor models that mirror their real-world counterparts is now open to the world's cancer researchers, with an international team of scientists developing the most comprehensive compendium to date of cell line and organoid cancer models paired with their original human tumors. Cell line models are cancer cells generally grown in flat layers, while organoids are three-dimensional clusters of cells suspended in gel that mimic the structure and behavior of tumors in the body, and the closer a model reflects the original tumor's biological features, the more reliable it is as a research tool. "One of the biggest challenges in cancer research is developing models that retain characteristics of the original tumors and behave in the culture dish like they do in the body. The new compendium addresses this by providing rigorously developed cell line and organoid models that have been deeply analyzed to ensure close fidelity to their source cells," said Van Andel Institute's Peter W. Laird, Ph.D., a co-author of the study. "The take-home message is that these models are excellent representations of their original tumors."

• A new approach tricks the immune system into accepting a transplanted organ permanently, as University of Pennsylvania researchers used a cold virus engineered to carry the gene for the protein CTLA4Ig to treat donor livers in preservation solution before transplant. The technique works because two chemical signals are required to trigger T cells to attack foreign tissues: a primary signal at the major histocompatibility complex and a costimulatory signal, and without the second signal the T cell will not develop into a cytotoxic T cell that destroys the invader tissues. By blocking that costimulatory pathway at the time of surgery and the period immediately after, the T cells become permanently tolerant of the new tissue.

  ---

💡Products/tools of the week

• The first desktop app designed to both run and train AI models locally has officially launched. Unsloth released Unsloth Desktop as a free, open-source Tauri-based app for Mac, Windows, and Linux, eliminating the need for Python environments or terminal commands. The app lets users download a model and start chatting immediately, with support for MLX, diffusion image/video, audio, and GGUF models across CPU and multi-GPU setups including NVIDIA, AMD, Intel, and Apple Silicon. Unsloth included features like 50% more accurate self-healing tool calls, sandboxed code execution, private web search, deep research, RAG, and MCP, plus the ability to connect Claude Code and Codex to local LLMs. Users can train models 2× faster with 70% less VRAM, and securely deploy LLMs remotely via Cloudflare HTTPS.

• A platform built specifically for AI agents has arrived to streamline website creation and management. Cactal provides a single API that lets AI agents build, host, and operate websites at scale, integrating with Claude Code, ChatGPT Codex, Cursor, Gemini, and any MCP-compatible agent. The platform gives agents full control over pages, CMS, domains, analytics, and versioning. Sites are built in real React code and hosted on a global edge network, with a nine-point quality control pipeline ensuring quality before publishing. Cactal supports use cases ranging from solo developers to agencies managing thousands of client sites, with built-in analytics, access controls, and SEO tooling all operable via agent commands.

• A competitor research tool has hit the market to help businesses understand exactly how rival websites grew from scratch. Tracetify allows users to paste a competitor URL and receive a detailed, sourced timeline covering everything from the site's first public mention through its key growth milestones. Each report includes launch events, early traction moves, and a verdict on which strategies remain replicable today. The tool draws from 12 data sources and includes over 100 linked references pointing to dated, publicly verifiable evidence. Tracetify offers two free traces before sign-up, with additional credits available as one-time purchases or through an unlimited subscription.

• A new open-source feedback management platform has arrived to give teams an AI-native alternative to tools like Canny. FeedLog lets teams collect and vote on feature requests, manage an interactive roadmap, and publish a standalone changelog, all within a single deployable package. The platform comes with built-in AI features including semantic duplicate detection and an AI changelog assistant that generates release notes in the user’s brand voice. FeedLog runs on Cloudflare Workers, Vercel, or Docker with a standard PostgreSQL backend, and is licensed under MIT with no tracked-user or seat limits.

  ---

• CNET: Meet the Flying, Swimming Robot Inspired by Aquatic Birds | What The Future

• NVIDIA: Accelerating Humanoid Robot Development With NVIDIA Isaac GR00T

• Fireship: I spent 3 days at MIT... the robot hype is worse than you think

Welcome to this week's edition of Rushing Robotics! What a week. We watched a lightbulb filament — yes, the same principle Edison patented — get reimagined as a decarbonization powerhouse, saw humanoid robots graduate from demos to mass-production plans, and tracked a quantum memory going live inside a real commercial network for the first time. If this edition proves anything, it's that the gap between "lab curiosity" and "industrial reality" is closing faster than ever.

🤯 Mind-Blowing

Researchers at NUS turned the humble incandescent filament into a compact reactor that drives industrial chemistry past 1,200°C using nothing but electricity — producing hydrogen from ammonia, recycling plastics, and converting methane all in one device. Meanwhile, South Korea's ASTRA-Net is spotting lung airways that human annotators missed entirely, redrawing the maps surgeons rely on for bronchoscopy. Add in a breakthrough explanation of how skin allergies escalate into body-wide anaphylaxis, Mitsubishi Motors committing an entire Kyoto production line to churn out 1,000 humanoid robots a month by 2027, and ORNL's cryogenic pellet tech officially licensed to accelerate stellarator fusion — and you have a week that reads like science fiction on fast-forward.

🔊 Industry Insights & Updates

The infrastructure layer of the future got serious this week. IonQ and EPB are building the Tennessee Quantum Communications Research Center around the world's first commercial quantum memory embedded in a live network, backed by $15 million over five years. Montage Technology kicked off trial production of the industry's first CXL 3.2 memory expander controller, pushing data transfer rates up to 64 GT/s for AI data centers. And in a sign of how unconventional compute is going mainstream, Extropic signed a $75 million letter of intent with the U.S. Department of Commerce to manufacture thermodynamic AI chips on American soil. Even coal chemistry got smarter, with a quartz-particle visualization trick boosting tar yields by 8.5%.

🧬 BioTech

Biology keeps revealing itself as a network, not a machine. A striking new review in Genes & Diseases frames tumors as wired participants in the nervous system — hijacking nerves, growing new ones, and forging neural connections to fuel their own survival. On the operating-room floor, researchers tracked surgeons' stress in real time during robot-assisted procedures, identifying four physiological signals that could reshape console design and workload planning. And Insilico Medicine nominated its 32nd AI-designed preclinical candidate, ISM9077, a potential first-in-class treatment for dry AMD and other ocular diseases that outperformed current therapies in preclinical models — with reach extending toward neurodegeneration and inflammatory disease.

💡 Products/Tools of the Week

The agent economy keeps compounding. FutureSearch is selling verifiable forecasting as a service — priced per question and already topping tournament leaderboards. HiveMind turns every AI coding session into a searchable, replayable team asset. BenchGen benchmarks agents inside digital-twin simulations and converts the results directly into RL training data. And Amorphic Labs can now learn your product from a URL and generate fully personalized, narrated demo videos for every prospect on your list — no recording required

🎥 Video Section

Proof beats promises — this week's clips show humanoid capability crossing new thresholds. Watch Figure 03 climb a ladder completely autonomously (DPCcars), get a close look at Foundation Robotics' new robotic hand in testing, and see how San Franciscans can already hire humanoid robots to clean their homes (ABC7 News Bay Area). The demos are no longer staged; they're operational.

Weeks like this one make it hard not to feel optimistic. A lightbulb filament is decarbonizing heavy industry, robots are moving from trade-show stages onto factory floors, and quantum memories are humming inside live networks. Each of these was a speculative headline just a few years ago — now they're production lines, licensing deals, and clinical tools. The future isn't arriving all at once; it's arriving in exactly these small, compounding steps, and we're lucky enough to watch it happen in real time. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A lightbulb-inspired reactor was unveiled by researchers at the National University of Singapore, who developed a compact device that uses a thin metal filament heated by electricity to drive industrial chemical reactions at temperatures above 1,200°C. The team, demonstrated the reactor in three decarbonisation applications: producing hydrogen from ammonia with near-complete conversion, recycling common plastics into chemical building blocks, and converting methane into valuable chemicals while co-producing hydrogen. The technology runs on electricity, making it compatible with renewable energy sources, and concentrates heat only in the filament rather than heating the entire reactor.

• A new AI framework called ASTRA-Net was developed by researchers at Pusan National University in South Korea to identify previously overlooked lung airway branches and generate more complete airway maps for bronchoscopy navigation. The framework was designed to address the challenge of incomplete airway maps reconstructed from CT scans, which can cause AI systems to overlook clinically important airways needed to reach deep lung lesions. ASTRA-Net uses a multi-stage deep learning architecture with an attention mechanism that focuses on regions where small peripheral airways are difficult to distinguish or may have been omitted from annotations.

• A key mechanism explaining how allergic inflammation in the skin can progress to a body-wide allergic response was uncovered by researchers from Tokyo University of Science and Kyoto University in Japan. The researchers found that IL-13, an immune signaling protein involved in allergic inflammation, acts on type 2 classical dendritic cells called cDC2, enhancing their ability to present allergens to other immune cells in a process known as licensing. This promotes the production of high-affinity IgE antibodies that can trigger systemic anaphylaxis. The team developed a murine model mimicking the atopic march and discovered that CX3CR1-positive cDC2 cells circulate in the bloodstream and transport allergens to secondary lymphoid organs such as the spleen. When the researchers blocked CX3CR1 with a drug, migration of these cells was severely impaired and production of high-affinity IgE antibodies dropped dramatically, suggesting this pathway as a promising therapeutic target for interrupting the atopic march.

• A plan to mass-produce AI-powered humanoid robots was announced by Mitsubishi Motors, which will use an unused engine production line at its Kyoto plant to manufacture robots developed by Tokyo-based startup Highlanders Inc. Production is expected to begin in the first half of 2027, with the facility capable of producing up to 1,000 robots per month. The robots stand approximately 1.75 meters tall and feature artificial intelligence that enables them to assess their surroundings autonomously, along with articulated hands capable of grasping and carrying objects. Mitsubishi will initially deploy the robots at its own manufacturing facilities to help address anticipated labor shortages before potentially offering them for sale to other manufacturers. The automaker said it will leverage the manufacturing expertise and quality systems developed through decades of vehicle production to support robot manufacturing.

• A licensing agreement was signed on August 5, 2026, transferring four cryogenic pellet fueling technologies from Oak Ridge National Laboratory to Type One Energy. The deal was finalized at the Nuclear Opportunities Workshop in Knoxville, Tennessee, where ORNL Director Stephen Streiffer and Type One Energy's Senior Vice President Matt Miles put pen to paper. These technologies create solid hydrogen pellets at near absolute zero temperatures and fire them into fusion plasma at high speed, a method that fuels plasma cores more effectively than gas puffing. The licensed inventions include a tritium-compatible gas gun, a cryogenic screw extruder, a pressure relief device, and a recirculation system that recycles excess extrusion into gas. Type One Energy, an Oak Ridge-based fusion company developing next-generation stellarator technology, will use these tools to accelerate commercialization of fusion energy.

🔊 Industry Insights & Updates

• A simple visualization technique using quartz particles revealed that reactor geometry can significantly improve coal tar yield and quality, according to a study published on May 27, 2026, in Sustainable Carbon Materials. Researchers from Chongqing University and Monash University placed inert quartz particles inside a fixed-bed reactor and observed that their surfaces darkened when exposed to tar-rich vapors, creating visible color patterns that traced vapor movement. The team, led by corresponding author Erfeng Hu, compared conventional reactors with modified designs containing four heating plates and a central gas-gathering tube, finding that the redesigned reactor increased total tar yield by approximately 8.5 percent and raised the proportion of light tar to 74.42 percent. The internal structures redirected vapors toward cooler regions, reducing unwanted secondary reactions that degrade valuable compounds.

• A partnership was announced on August 3, 2026, between IonQ and EPB to launch the Tennessee Quantum Communications Research Center in Chattanooga. The facility will house the world's first commercial Quantum Memory embedded within an operational communications network, staffed by IonQ scientists. IonQ committed $15 million over five years to fund the initiative, while EPB will provide the live network development test bed and serve as the primary commercialization partner. The center aims to accelerate development of the Quantum Internet by enabling quantum computers to reliably share information across long distances.

• Trial production of the industry's first CXL 3.2 Memory eXpander Controller chip was announced on July 31, 2026, by Montage Technology, a semiconductor company based in Hengqin, Guangdong, China. The chip, called MXC, is designed to address growing demands for memory expansion and resource pooling in AI, cloud computing, and data center environments by enabling customers to build memory infrastructure with higher capacity, improved resource utilization, and greater flexibility. The CXL 3.2 MXC complies with CXL Type 3 specifications and supports PCIe 6.x and CXL 3.2 protocols, delivering data transfer rates of up to 64 GT/s, with dual DDR5 memory controllers supporting up to 8000 MT/s DDR5 memory. Stephen Tai, President at Montage Technology, said the trial production marks an important milestone in the company's pursuit of CXL innovation and commercialization.

• A $75 million letter of intent was signed between Extropic and the U.S. Department of Commerce to accelerate development and domestic manufacturing of thermodynamic computing hardware for AI. The non-binding agreement, funded through the CHIPS Research and Development Office, would support development of Extropic's Thermodynamic Sampling Units (TSUs), probabilistic semiconductor chips designed to improve energy efficiency of generative AI, robotics, and simulations. The project includes scaling production of the Z1 processor, developing rack-scale thermodynamic computing systems, and establishing U.S.-based manufacturing for a future Z1.5 chip using mature semiconductor process nodes. Extropic founder and CEO Guillaume Verdon said the partnership with the Trump Administration, Secretary Lutnick, and Executive Director Frauenhofer will help manufacture this new category of compute on American soil.

Subscribe now

🧬 BioTech

• A new review article published in Genes & Diseases reveals that tumors are not isolated masses but wired participants in a living communication network with the nervous system. Nerves send signals that fuel cancer cell growth, drive metastasis, and shield tumors from treatment, while tumors strike back by rewiring nearby nerves to serve their own survival. Three processes make this possible: tumors invade nerve structures, trigger new nerve fiber growth, and even forge entirely new neural connections, all linked to more aggressive disease and worse outcomes. The review, authored by a research team from Chongqing Medical University, positions the nerve-tumor axis as a concept that could transform how cancer is treated.

• A study revealed how surgeons’ stress changes moment by moment during live robot-assisted operations, using lightweight sensors to track brain, muscle, and heart activity. Seven experienced urologists wore sensors recording electroencephalography, surface electromyography, and electrocardiography while performing robot-assisted procedures, and after each surgery they watched recordings of their console view and marked moments when their stress changed. The team analyzed 151 stress-rated moments and found four physiological signals reliably tracked stress: brain activity linked to mental arousal increased, upper shoulder muscle tension increased, heart rate sped up, and natural variability between heartbeats decreased. Professor Shimomura said the approach could help identify high-demand operative moments during postoperative review and serve as a way to evaluate the robotic surgery work system, including console ergonomics, communication routines, and workload distribution.

• A new preclinical candidate named ISM9077 was nominated by Insilico Medicine on August 6, 2026, as a potential first-in-class Target Y inhibitor for treating ocular diseases including dry age-related macular degeneration, uveitis, and dry eye disease. The candidate was designed, evaluated, and optimized using Chemistry42's integrated generative AI models, making it the company's 32nd preclinical candidate nominated since 2021. In preclinical studies, ISM9077 outperformed current therapies in dry AMD, uveitis, and dry eye models while demonstrating a favorable safety profile with a wide safety margin. The compound exhibits good permeability and excellent retinal tissue exposure, supporting potential delivery through both oral administration and eye drops. Beyond ocular diseases, the novel mechanism modulates pathological inflammation, supporting broad indication expansion into aging-related and other inflammation-related disorders including neurodegenerative diseases, inflammatory bowel disease, cardiovascular conditions, MASH, and obesity.

  ---

💡Products/tools of the week

• An AI-powered forecasting platform now predicts the future with verifiable accuracy. FutureSearch built a system that produces probability, numeric, date, categorical, conditional, and decision forecasts, drawing on a shared world model that compounds with use. The platform charges $0.15 per question at low effort and $2 at high effort, with categorical, thresholded, and conditional forecasts always requiring high effort. For deeper analysis, FutureSearch supports multi-agent research teams that run parallel web searches, costing $0.30 to $2 per question. The company maintains a verified public track record, ranking first in Metaculus's Summer 2026 FutureEval tournament, 17th of 304 models on ForecastBench, and holding the best pooled score on its 1,907-question pastcasting benchmark.

• An engineering analytics platform launched. HiveMind built a system that captures every AI coding agent session into a shared, searchable record so developers can replay, fork, and continue past runs instead of starting over. A lightweight local daemon discovers sessions, scrubs secrets on-device, and syncs logs to a central backend where tool calls, diffs, and token costs are indexed. Teams gain visibility into what agents shipped, what stalled, and where token spend went, enabling continuous improvement of AI-assisted workflows.

• Benchmarking AI agents just moved into digital-twin simulations. BenchGen built a platform that runs agents inside these environments to capture full decision trajectories and verify real-world performance. It converts benchmark runs into reinforcement learning training data compatible with PPO, GRPO, and PRM methods, enabling continuous agent improvement. BenchGen supports air-gapped and sovereign deployments for government, defense, and regulated industries. The platform serves over 500 teams, has evaluated more than 1,400 agents, and has captured over 2 million trajectories across 750-plus RL environments.

• Personalized product demo videos can now be generated without manual recording. Amorphic Labs built an AI-powered platform that uses autonomous agents to create tailored demo videos for B2B sales teams. Users provide a list of prospects and the agent researches each one, seeds the product with prospect-specific data, records a tailored workflow walkthrough, and outputs a narrated, polished video ready to send. The agent selects relevant workflows per account and addresses each prospect by name, ensuring no two demo videos are alike. It can learn a new product from a URL and sandbox login on the same day, requiring no manual recording from the sales team.

  ---

• DPCcars: Figure 03 Robot Just Climbed a Ladder Completely on Its Own

• Foundation Robotics: Testing Our New Robotic Hand

• ABC7 News Bay Area: You can hire these humanoid robots to clean your home in San Francisco

Welcome to this week's edition of Rushing Robotics! As we close out July 2026, the pace of breakthroughs feels less like incremental progress and more like a cascade of paradigm shifts hitting simultaneously across robotics, biotechnology, quantum computing, and materials science. Whether it's robots that think ahead while they act, mitochondria doubling as immune warriors, or single-electron memory chips that could put large language models in your pocket—the future isn't just arriving. It's flooding in.

🤯 Mind-Blowing

This week's developments remind us that the most profound breakthroughs often come from rethinking fundamentals—how robots plan motion, how our cells defend us, and how quantum mechanics can decode cancer's secrets. From MIT's VLASH technique that eliminates robotic "thinking pauses" to a tissue-like 3D-printable material that can filter molecules and grow organs, these stories push the edges of what we thought possible.

🔊 Industry Insights & Updates

The industry pulse this week beats with historic firsts and commercial leaps. Google DeepMind's Gemini Robotics 2 brings whole-body intelligence to humanoids, Zoox secures the first-ever NHTSA exemption for purpose-built robotaxis to carry paying passengers, and Fudan University's single-electron memory chip could collapse the data storage equation by 200,000x. Meanwhile, Nagoya University's skeletal editing of nanocarbons opens an entirely new materials frontier.

🧬 BioTech

In biotech, the body's own mechanisms are proving to be the most powerful tools we have. From focused ultrasound opening the blood-brain barrier for glioma treatment, to a gut bacterium metabolite that repairs HIV-related intestinal damage, to a natural protein called BMP3 that can reverse pulmonary arterial hypertension—this week's research turns the immune system, mitochondria, and even our microbiome into allies against diseases once thought intractable.

💡 Products/Tools of the Week

For builders and makers, this week brings a fresh toolkit: MySpec for AI-driven technical specifications, Context.dev for web scraping built for AI pipelines, Artificial Analysis for independent LLM benchmarking, and AICanRun for checking which local LLMs your phone can actually handle.

🎥 Video Section

This week's video lineup showcases the future in motion—Neuralink's telepathic wheelchair control, Google DeepMind and Apptronik's Gemini Robotics 2 in action, and the FLUX-mimic collaboration with Black Forest Labs. Watch the breakthroughs unfold.

The convergence happening across disciplines is unlike anything seen before—quantum computing meeting cancer immunotherapy, robotics meeting whole-body AI, materials science meeting energy and computing simultaneously. We're entering an era where fields don't just advance in parallel; they amplify each other, compressing decades of progress into years. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A new method called VLASH emerged as MIT researchers found a way to make robots faster by helping them think ahead while they act. The system enables a robot's AI model to predict its future position and use that information to plan the next set of motions while completing its current action chunk, eliminating the pauses that typically occur between movements. MIT researchers led by Song Han, an associate professor in the Department of Electrical Engineering and Computer Science, developed the technique alongside co-lead authors Jiaming Tang and Yufei Sun, with collaborators from Nvidia, UC Berkeley, UC San Diego, and Caltech. The method doubled the speed of robots performing pick-and-place tasks and boosted performance in dynamic activities like table tennis and Whack-a-Mole, all without adding computational overhead to the planning process.

• A surprising immune role for mitochondria came to light as University of Queensland researchers led by Ronan Kapetanovic revealed that the cell's energy factories actively fight bacterial infections by splitting into smaller units. The team, working with collaborators in Switzerland and Spain, found that E. coli infections trigger mitochondrial fission in mouse and human macrophages as well as in Caenorhabditis elegans worms, boosting bacterial clearance. Fission activates the mitochondrial unfolded protein response and drives production of antimicrobial lipid droplets through the transcription factor ATF5, which also curbs excessive defense to restore cellular balance. Published in Science Immunology, the study suggests targeting this pathway could help fight antibiotic-resistant pathogens, especially since inhibiting the enzyme HDAC6 promoted fission and enhanced clearance of Salmonella typhimurium, a pathogen that normally hijacks mitochondria to survive.

• A breakthrough in cancer immunotherapy prediction took shape as Cleveland Clinic and IBM researchers developed Q-CHIPP, a quantum convolutional neural network that predicts which tumor mutations will activate the immune system to attack cancer cells. The framework tackles one of the hardest problems in immuno-oncology by identifying immunogenic neoantigens, the abnormal proteins on cancer cell surfaces that signal the immune system to respond. The model outperformed traditional computing methods under similar constraints and parameter limits, and it can learn from training sets as small as 150 samples, overcoming the limitations of small datasets like Cleveland Clinic's 196-candidate neoantigen atlas published in 2024. Senior author Tyler Alban of Cleveland Clinic said the creation exemplifies team science, with experts in cancer, computational sciences, and quantum computing all learning from one another. The team also achieved a significant technical milestone by scaling the approach to full-length peptide modeling on quantum hardware using 46 qubits.. The researchers plan to continue enhancing the model to enable more accurate identification of therapeutic targets, potentially accelerating the development of personalized cancer immunotherapies.

• A versatile new 3D-printable material emerged as University of Texas at Austin researchers developed a tissue-like substance that can filter molecules, grow new organs, and recover critical minerals from wastewater. Published in Nature Materials, the work was led by Manish Kumar, a professor in the Cockrell School of Engineering, with student Aida Fica driving the breakthrough after years of slow formation and instability challenges. Fica solved the problem by bringing together two oils with different solubilities to form droplets, then jamming them together with a centrifuge to create large, tissue-like materials in just minutes. The flexible material can be customized with proteins to conduct ion currents like nerve tissue, opening possibilities for brain-modeled computing, and can filter ammonium from oil and gas wastewater. The researchers are now adapting the technology to extract lithium and rare-earth elements through a project with the U.S. Department of Energy's Advanced Research Projects Agency-Energy.

• A breakthrough in machine communication unfolded as Cornell researchers revealed their "microwave brain" microchip can now encode messages into its own private language using microwave token embeddings. Bal Govind, who led the study in Alyssa Apsel's laboratory at the Cornell Duffield College of Engineering, showed that the low-power chip can convert information such as navigation commands into distinctive microwave pulses that only another matching neural network can interpret. The research, published July 29 in Nature Communications, also found that feeding gigabit-per-second data into the chip naturally generates probabilistic bits, allowing the team to reconstruct a satellite image of a tropical storm while reducing transmitted data by about eightfold. The technology could offer a new form of hardware-based cybersecurity and help small satellites transmit images back to Earth despite power and bandwidth limitations.

🔊 Industry Insights & Updates

• A major leap in robotics arrived as Google DeepMind unveiled Gemini Robotics 2, bringing whole-body intelligence to robots for the first time. The new system enables humanoids to walk, crouch, stretch, and manipulate objects while teaming up with other robots. Three models power the update: Gemini Robotics 2 for motor control, Gemini Robotics ER 2 for reasoning and planning, and Gemini Robotics On-Device 2 for local operation. Google DeepMind designed the platform to adapt to entirely new robotic bodies within hours, marking a significant step toward general-purpose physical AI.

• A way to build nanocarbons emerged as Nagoya University researchers used skeletal editing to cut and reform bonds inside flat carbon molecules, reshaping their interiors for the first time. The team created chiral nanocarbons that exist in mirror-image forms, including structures with 10-carbon rings and a double-helix motif. The molecules emit light in spiral patterns and can store electrical charges without breaking down. One maintains its handedness up to 280°C, while another forms porous crystals that trap and release carbon dioxide. Published in Nature Communications on July 28, 2026, the work extends skeletal editing from pharmaceuticals into carbon nanostructures and could advance gas storage and ultra-low-power electronics.

• A breakthrough in semiconductor technology emerged as a Fudan University team developed a chip that stores data using a single electron, collapsing the number needed from roughly 200,000 to just one. Microelectronics professor Zhou Peng and his colleagues detailed the working 2D flash memory, named Guiyi, in a paper published in Science on July 16. The device uses a graphene layer to trap electrons and amplify their signal from tens of millivolts to a robust 0.5 volts, a tenfold improvement over previous single-electron attempts. The technology could enable large language models to run on mobile phones using minimal power without losing memory during conversations. Zhou plans to set up a company later this year and aims to commercialize the technology in three to five years.

• A historic first happened as the National Highway Traffic Safety Administration granted Amazon-owned Zoox a temporary exemption to commercially deploy its purpose-built robotaxis without traditional human controls. The exemption covers up to 2,500 vehicles annually over two years, allowing Zoox to charge passengers for rides for the first time. Zoox will begin paid service in Las Vegas soon, with additional markets following as state-level requirements are completed. The company's custom-built vehicles lack steering wheels, brake pedals, and other conventional driver controls, which previously conflicted with eight Federal Motor Vehicle Safety Standards. Zoox CEO Aicha Evans called the decision a milestone for both the company and autonomous mobility more broadly.

Subscribe now

🧬 BioTech

• A promising advance in brain tumor treatment emerged as UVA Health researchers found that gliomas may be even more receptive to focused ultrasound drug delivery than normal brain tissue. The team, led by Richard Price, PhD, co-director of UVA's Focused Ultrasound Cancer Immunotherapy Center, used tiny microbubbles activated by sound waves to open the blood-brain barrier so drugs can enter exactly where needed. Working with Dr. Wilson Miller and graduate student Matthew Hoch, Price developed an MRI approach that allowed drug-delivery measurements with unprecedented resolution. Published in Radiology, the findings allay concerns that brain tumors might resist the cutting-edge approach and identified a Goldilocks size range where drug molecules worked best. The research could benefit patients with glioblastomas, the deadliest form of brain cancer, which kills more than 10,000 people in the United States every year.

• A promising treatment for HIV gut damage arrived as UC Davis School of Medicine researchers discovered that a metabolite produced by beneficial gut bacteria can repair intestinal harm caused by the virus and significantly improve antiretroviral therapy. The metabolite, 10-hydroxystearic acid or 10-HSA, comes from Lactiplantibacillus plantarum, a common lactic acid bacterium found in over-the-counter probiotics and fermented foods. Senior author Satya Dandekar, a distinguished professor in the Department of Medical Microbiology and Immunology, and first author Dylan Kramer published the findings in Nature Microbiology on July 30, 2026. Using X-ray crystallography, the team showed that 10-HSA binds to PPAR-alpha, triggering epigenetic changes that restore mitochondrial function and rebuild the gut barrier. In two nonhuman primate studies, 10-HSA alone repaired intestinal damage and improved gut function, while combining it with ART led to faster viral clearance and restored beneficial gut microbiota. The University of California has filed a patent application related to the therapeutic strategy.

• A new treatment hope emerged as Fralin Biomedical Research Institute at VTC researchers identified a natural protective protein called BMP3 that could reverse pulmonary arterial hypertension. The team, led by senior author Yassine Sassi and first author Aymen Halouani, found that BMP3 levels drop significantly in patients with the disease, causing lung blood vessels to thicken and narrow. By restoring BMP3 through both recombinant protein and gene therapy in preclinical models, the researchers reversed key features of the disease, including damaged blood vessels and impaired heart function. Published in the European Respiratory Journal, the study marks the first time BMP3 has been linked to pulmonary arterial hypertension, and the researchers have filed a patent application covering BMP3-based therapies.

  ---

💡Products/tools of the week

• A AI-powered platform launched to transform how developers create technical specifications. MySpec interviews users like a senior architect and produces a structured four-file bundle with acceptance criteria and diagrams. Users can refine specs conversationally, export them as ZIP files, or connect directly to coding agents like Cursor and Claude Code. The platform supports both new and existing projects and integrates with popular tools for seamless handoff.

• Web scraping for AI development just got more powerful. Context.dev launched as a data API built for teams creating AI agents and LLM pipelines, capable of scraping any URL into clean markdown or HTML. The platform crawls entire websites via sitemap discovery and extracts structured data using a provided JSON schema. A Brand Intelligence API retrieves logos, colors, fonts, and company metadata from any domain. Every request includes JavaScript rendering, anti-bot handling, and stealth proxies, with typed SDKs available for TypeScript, Python, and Ruby.

• Independent benchmarks and analysis of AI language models and API hosting providers are now available through Artificial Analysis. The platform evaluates models across key performance metrics including intelligence, output speed, latency, and cost per task. It features leaderboards for text, image, video, speech, and coding agents, along with capability indices for industries like finance, legal, and healthcare. Users can access personalized model recommendations, custom visualizations, and expanded benchmark data through premium plans.

• A compatibility checker now tells smartphone users which local large language models their devices can handle. AICanRun analyzes chip, RAM, and memory bandwidth across 132 phones, 51 models, and 294 quantization variants. It uses transparent RAM math accounting for weights, KV cache, and runtime overhead. Each supported model comes with recommended quantization level, estimated inference speed in tokens per second, and the right app for deployment. No installation needed.

  ---

• mimic: mimic x Black Forest Labs: Introducing FLUX-mimic

• Neuralink: Telepathic Wheelchair Control | Neuralink

• Google DeepMind and Apptronik: Gemini Robotics 2 brings whole body intelligence to robots

Welcome to this week's edition of Rushing Robotics! What a week. From brain-controlled robots to quantum teleportation outperforming classical transmission, the pace of innovation isn't just accelerating—it's feeding on itself. This week alone, we're watching open-source AI shatter parameter records, diamond-based quantum computers ship without cryogenics, and CRISPR reveal itself as something far more sophisticated than the precision scissor we once imagined.

🤯 Mind-Blowing

BrainCo launched a brain-to-robot AI platform that turns neural signals into robotic actions—affordable and commercially viable. Moonshot AI dropped Kimi K3, a 2.8-trillion-parameter open-source model beating OpenAI and Anthropic on coding benchmarks. Biren Technology unveiled optical "supernode" interconnects that blow past the 128-GPU copper limit. Researchers from Beijing Institute of Technology published a framework for autonomous microrobots navigating the human body for drug delivery. And the University of Science and Technology of China proved quantum teleportation beats direct photon transmission by nearly 3x in lossy channels—the first head-to-head test ever.

🔊 Industry Insights & Updates

The infrastructure layer is getting loud. MIT printed a 2.3-meter concrete bridge in 30 minutes that held over 2,000 pounds. AGIBOT launched four new robots at WAIC 2026—humanoids, industrial platforms, and a direct-drive manipulation system. SAXON Q shipped diamond-based quantum computers running at room temperature—128 and 512 qubits, no cryogenics, plugging into standard server racks. Bristol Myers Squibb deployed their second NVIDIA DGX SuperPOD, giving every scientist access to what used to be a small-team luxury. What required specialized facilities is becoming rack-mountable and accessible.

🧬 BioTech

CRISPR just got an upgrade in how we understand it. Prof. LI Ming's team discovered CRISPR-Cas doesn't just snip viruses—it commands a network of 20+ innate immune modules called "CRISIS." St. Jude researchers found sickle cell disease ages blood stem cells prematurely, starting as early as age 6, and senolytic drugs can reverse the damage. University of Virginia scientists developed a nanoparticle + focused ultrasound approach to deliver microRNAs through the blood-brain barrier to fight glioblastoma. Biology is being rewritten at the immune, cellular, and molecular levels simultaneously.

💡 Products/Tools of the Week

Four tools worth your attention. PolyTalk delivers open-source, self-hosted speech-to-speech translation across 30+ languages with sub-second latency—no cloud required. Scribe automatically extracts decisions and patterns from git repos and coding agent sessions into a searchable local knowledge base. Zro offers private inference for open-weight coding models like GLM-5.2 and MiniMax M3 with zero request retention. Script.it brings a scripting-agent approach to workflow automation that's easier to verify than traditional no-code platforms.

🎥 Video Section

Three drops worth your screen time. Unitree Robotics showcases real-time omni-modal interaction driving whole-body mobile manipulation. Mimic introduces the M1 hand, a new benchmark in dexterous manipulation hardware. And Generative Bionics delivers a six-month update on GENE.01, their bionic system coming to life.

We're watching a dozen different fields hit escape velocity at the same time. Room-temperature quantum computers, brain-controlled robotics, CRISPR acting as an immune commander, teleportation beating direct transmission—these aren't incremental updates. The frontier isn't just moving. It's multiplying. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A major step forward in neurotechnology has been taken with the launch of BrainCo's new brain-to-robot AI platform, announced by the Chinese startup this week. The platform allows users to control robotic systems using only their brain signals, processed through advanced AI algorithms that interpret neural patterns and convert them into robotic actions. BrainCo designed the system to be more user-friendly and affordable than previous brain-computer interfaces, opening possibilities for applications in prosthetics, industrial automation, and assistive technology. The company believes this launch represents a crucial milestone in making brain-controlled robotics commercially viable and accessible to a broader market.

• A new world record for open-source AI has been set as Chinese start-up Moonshot AI unveiled Kimi K3, a model with 2.8 trillion parameters that surpasses all previous open-source offerings. The Beijing-based company said the model outperformed systems from OpenAI and Anthropic in certain evaluations, particularly in coding and software engineering tasks. Moonshot designed K3 for long-horizon coding, knowledge work, and reasoning tasks. The model will be available through Moonshot's Kimi chatbot and developer tools, with full weights released on July 27.

• A new optical technology for linking AI chips has been unveiled by Chinese chip start-up Biren Technology at the World Artificial Intelligence Conference in Shanghai. The company's "supernode" solutions use near-packaged optics to connect up to 1,024 processor cards across a single cluster, bypassing the 128-GPU limit of traditional copper-wired systems. Biren developed this approach to compete with Nvidia's platforms as AI models grow to trillions of parameters. The Shanghai-based firm projects mass commercial roll-out of the optical technology by 2028.

• A comprehensive review of autonomous microrobots for drug delivery has been published by researchers led by Professor Huaping Wang of the Beijing Institute of Technology, outlining how tiny robots can sense, navigate, and release drugs on demand inside the human body. The paper, published in Cyborg and Bionic Systems on June 29, 2026, introduces a four-tier hierarchy based on barrier traversal capability, ranging from local delivery in accessible cavities to immune-privileged sites like the brain. Co-authors Shihao Zhong, Anping Wu, Shanming Bai, Henwei Huang, Toshio Fukuda, and Huaping Wang describe how magnetic, acoustic, and biohybrid propulsion systems can transform passive drug carriers into active therapeutic robots, with early clinical entry likely through gastrointestinal, bladder, and eye applications.

• Quantum teleportation has been proven to beat direct photon transmission in long-distance communication, thanks to new research from the University of Science and Technology of China. The research team, led by co-senior author Chaoyang Lu, created an all-optical method to generate entangled photon pairs that largely avoid the losses suffered when photons travel directly through communication channels. When tested through a channel with only about 1% transmission capacity, teleportation proved nearly three times more efficient than direct transmission, even with optimal quantum cloning applied. Published in Nature Physics, the study marks the first experimental head-to-head comparison between teleportation and direct photon sending, moving the elegant four-decade-old concept closer to practical use in quantum relays, repeaters, and connected quantum processors.

🔊 Industry Insights & Updates

• A breakthrough in construction has been achieved as MIT researchers developed a framework that makes 3D-printed concrete structures more efficient and practical. The team designed and printed a 2.3-meter concrete bridge in just 30 minutes using off-the-shelf mortar, then load-tested it to hold more than 2,000 pounds with virtually no bending. Co-first authors Hajin Kim-Tackowiak and Zane Schemmer, working with senior author Josephine Carstensen, created a system that bakes printer limitations directly into the design optimization, generating fully printable designs in about two minutes on a laptop. The biggest surprise was that printer hardware, not concrete itself, limits efficiency. Their analysis revealed that a machine able to lay a 1-centimeter bead instead of the current 4 centimeters could reduce material use by 76 percent while staying within safety margins. The research, published in Additive Manufacturing and funded by the National Science Foundation, provides a roadmap for printer-makers to unlock large gains in efficiency.

• Four new robotics products have been unveiled by AGIBOT at the World Artificial Intelligence Conference 2026 in Shanghai, expanding its portfolio across humanoid platforms, industrial robots, and dexterous manipulation systems. The launches include the A3 Ultra full-size humanoid robot, X2 Edu, G2 Max heavy-payload industrial robot, and OmniHand 3 Ultra-M direct-drive manipulation system. More than 30 AGIBOT robots are operating across WAIC venues, providing visitor guidance and on-site support. Peng Zhihui, Co-founder, President and CTO of AGIBOT, said the industry should judge embodied AI by whether robots can be reliably integrated into real operating environments, not by one-time demonstrations.

• A new line of diamond-based quantum computers operating at room temperature has been launched by German start-up SAXON Q, marking the first time nitrogen-vacancy center processors have exceeded ten physical qubits. The company released two models, the SXQ128 with 128 qubits and the SXQ512 with 512 qubits, both designed to fit into standard server racks and run on ordinary AC wall outlets without cryogenic cooling or vacuum systems. Spun out of Leipzig University and led by CEO Prof. Dr. Marius Grundmann and CTO Prof. Dr. Jan-Berend Meijer, SAXON Q solved a long-standing manufacturing bottleneck through a proprietary sulfur co-implantation process that achieves over 85 percent conversion yield, far surpassing traditional rates of 1 to 10 percent. The SXQ128 is available to order immediately with deliveries starting late summer 2026, while the SXQ512 ships in Q2 2027.

• A second NVIDIA DGX SuperPOD has been deployed by Bristol Myers Squibb, built on eight DGX Vera Rubin NVL72 systems that the company calls the most powerful and energy-efficient AI cluster in life sciences. The new system delivers up to 10x the performance per megawatt of the infrastructure it replaces, giving every BMS scientist access to a unified AI platform rather than just a small group of researchers. Erin Davis, vice president of research business insights and technology at BMS, said the system opens up limitless compute to all scientists, enabling faster drug discovery cycles across the full pipeline.

Subscribe now

🧬 BioTech

• A breakthrough in understanding bacterial immunity has been achieved as researchers revealed CRISPR-Cas does far more than act as a lone sniper against viruses. Prof. LI Ming from the Institute of Microbiology of the Chinese Academy of Sciences led a study showing that CRISPR-Cas serves as a commander that orchestrates a network of innate immune systems. The team discovered more than 20 innate defense modules embedded within type I CRISPR-Cas loci, controlled by small crRNA-like RNAs that guide CRISPR machinery without triggering DNA cleavage. These CRISPR-supervised systems, termed “CRISIS,” are kept in check when CRISPR is active to avoid fitness costs. However, when phages deploy anti-CRISPR proteins to suppress CRISPR, the brakes are released and CRISIS systems unleash innate immunity. Published in Nature on July 22, 2026, the findings reshape the view of bacterial defense.

• A new study has revealed that sickle cell disease causes premature aging of blood stem cells, a problem that may be reversible with anti-aging drugs. Researchers at St. Jude Children's Research Hospital found that blood stem cells from young patients as early as 6 years old showed markers of senescence, making them less capable of producing healthy blood cells. When the team treated mice implanted with patient stem cells using senolytic drugs, the damaged cells were eliminated and blood-forming potential was restored. The study, published in Science Translational Medicine on July 22, 2026, was led by senior co-corresponding author Shannon McKinney-Freeman, first author Aditya Barve, and co-corresponding author Akshay Sharma, who believe the findings could improve both stem cell collection and outcomes for sickle cell gene therapy.

• A new experimental approach to targeting glioblastoma, the deadliest brain cancer, has been developed by scientists at the University of Virginia Comprehensive Cancer Center. The team, led by Roger Abounader, MD, PhD, identified microRNAs that can simultaneously suppress multiple malfunctioning genes responsible for the tumor's formation and growth. The researchers use brain-penetrating nanoparticles, focused ultrasound waves, and microbubbles to deliver the miRNAs through the blood-brain barrier, which typically blocks treatments from reaching the brain. Lab testing showed the approach slowed tumor growth and extended survival in animal models. The findings were published in the Journal of Clinical Investigation.

  ---

💡Products/tools of the week

• Live audio translation now runs entirely on your own infrastructure with sub-second latency. PolyTalk created this open-source, self-hosted speech-to-speech platform supporting over 30 languages. It works with microphones, browser tabs, meetings, and audio streams without sending data to external cloud services. The platform comes in a free Community Edition and a paid Enterprise Workspace edition, both fully on-premise. PolyTalk built it for healthcare, government, legal, education, and other fields needing private multilingual communication

• Extracting decisions, patterns, and learnings from git repositories, Claude Code and Codex sessions, and self-sent links into a curated, semantically searchable markdown knowledge base can now happen automatically. Scribe developed this single-binary CLI that runs on a cron schedule and works fully local, with no external vector database or hosted infrastructure required. The tool uses BM25 full-text search over auto-generated wiki files, ensuring lookups stay fast and cheap. Scribe integrated it with AI coding agents like Claude Code and Codex by writing a handshake block into their config files, enabling every agent session to query the knowledge base before making architectural or library recommendations.

• Open-weight coding models can now run through a private inference endpoint with zero request retention and no training on customer data. Zro built this endpoint to support popular coding agents including Claude Code, Codex CLI, Cursor, Cline, and others via an npm package and OpenAI- or Anthropic-compatible APIs. The platform is optimized for long-context, multi-turn coding sessions using HyperQuant compression and custom attention kernels on AMD, NVIDIA, and TPU hardware. Zro currently offers MiniMax M3 and GLM-5.2, with additional open-weight models planned. Billing starts at $20 per month for $60 of inference spend, and Zro also offers pay-as-you-go usage packs.

• A new AI-powered automation platform now lets users build reliable, verifiable workflows using a scripting agent approach. Script.it developed this platform to be easier to build and verify than traditional no-code tools like n8n, Zapier, or Make, along with visual workflow builders and black-box AI agents. Users can define structured scripts with inputs, steps, and integrations that run automatically or on a schedule. Script.it also supports human-in-the-loop and agent-in-the-loop patterns, making it suitable for recurring tasks such as data gathering, digest compilation, and email delivery.

  ---

• Unitree Robotics: Real-Time Omni-Modal Interaction Driven Whole-Body Mobile Manipulation

• mimic: Introducing the mimic hand M1

• Generative Bionics: [Six-Month Update]: GENE.01 Comes to Life

Welcome to this week's edition of Rushing Robotics! This issue is stacked: an open-weights behemoth just dethroned the closed-source kings, Stanford may have found the aging "off switch," robots are quietly passing their factory exams with near-perfect scores, and a new solar cell record is bending the curve on clean energy. Whether you're here for the algorithms, the actuators, or the biology — we've got something that'll make you pause and stare.

🤯 Mind-Blowing

The breakthroughs that made us whisper "no way" under our breath. This week, the future arrived in 2.8-trillion-parameter open models, single-shot 3D printing, mechanical soft sensors with zero electronics, and a quantum primitive that could rewrite what quantum computers can do. These aren't incremental updates — they're tectonic shifts happening in real time.

🔊 Industry Insights & Updates

The pulse of the industry — deployments, records, and the companies turning demos into reality. Humanoid robots are stepping out of labs and onto production lines, solar cells just smashed a new efficiency record, and embodied AI is getting a massive perception upgrade. The gap between "promising prototype" and "shipping product" is closing fast.

🧬 BioTech

From cancer’s repair mechanisms to the immune architects fighting tumors — the frontier where biology meets engineering. This week brings a new class of drugs that strips cancer cells of their self-healing superpowers, the discovery of the immune cell that orchestrates anti-tumor attacks, and a surgical breakthrough that delivered years of pain relief where nothing else worked.

💡 Products/Tools of the Week

The sharpest new tools in the shed, handpicked for builders, marketers, and engineers. Whether you're cloning winning ad campaigns from a single chat, connecting AI agents to dozens of data sources through one API, generating UGC-style video ads at scale, or shipping apps faster with self-healing end-to-end tests — this week's lineup is built to save you hours and multiply your output.

🎥 Video Section

Because some breakthroughs deserve to be seen in motion. This week's video lineup features open-source full-body robot control, dexterous humanoid hands, a new general-purpose robot hitting the market, and machines that learn while they work. Press play — the future moves fast, and these clips move faster.

The future isn't a distant horizon — it's the headline of today's paper. Open models are rivaling closed giants, robots are learning to build our cars, and scientists are rewriting the rules of aging and cancer treatment. Each discovery accelerates the next, and we're all front-row witnesses to humanity's greatest acceleration. Stay hungry, stay futurish!

🤯 Mind-Blowing

• Moonshot AI has officially released Kimi K3, marking a watershed moment in artificial intelligence development as the first open-weights model to demonstrably match and exceed the capabilities of leading proprietary systems. Launched on July 16, 2026, this 2.8-trillion-parameter Mixture of Experts architecture establishes itself as the largest open-source model ever deployed while delivering performance metrics that rival—and in key benchmarks, surpass—closed-source competitors from Anthropic and OpenAI. Independent evaluation data supports this claim, with Kimi K3 achieving a 1679 Elo rating on Arena Frontend Code, securing the top position ahead of Claude Fable 5, and recording 94.2% accuracy on the Multi-Level Reasoning benchmark, edging out GPT-5.6's 93.1%. Technical specifications further distinguish this release through Moonshot's proprietary "Delta Attention" mechanism and a 1 million token context window, enabling comprehensive processing of extensive codebases and multi-document analyses without the segmentation constraints that have historically limited open-source alternatives.

• A single immune cell type has been identified as the main driver of organ aging, according to Stanford Medicine researchers who found that blocking one receptor on tissue-resident macrophages preserved the youthfulness of multiple organs in mice. The study, published in Science on July 16, 2026, was led by senior author Katrin Andreasson and lead author Jessy Tan, who showed that disabling the EP2 receptor on these macrophages restored their ability to clear out senescent neutrophils, the toxic zombie-like cells that accumulate with age and damage surrounding tissues. Mice lacking EP2 stayed leaner, stronger, and sharper, with reduced inflammation in the brain, heart, liver, kidney, and colon, and memory performance matching that of much younger animals.

• A new 3D printing method that creates solid shapes in a single shot has been demonstrated by researchers at the University of Utah's John and Marcia Price College of Engineering. The technique uses a nanoscale mask to diffract laser light into a holographic pattern, fusing material solid in about 20 seconds, a stark contrast to the hours required by other laser-based printing methods. Professor Rajesh Menon from the Department of Electrical and Computer Engineering led the study, along with lab member Dajun Lin, publishing their findings in Nature Communications on July 7, 2026. The mask acts like a nanopatterned lens that compensates for the substrate's diffraction, channeling laser energy only to the volume that becomes the final print. The team printed microtubule assemblies with diameters as small as 6 micrometers and dimensional ratios up to 120:1, demonstrating that the structures could transport liquid via capillary action and withstand compression tests.

• A fully mechanical soft sensor that turns touch directly into robotic action without any electronics has been developed by researchers at the National University of Singapore's College of Design and Engineering. The device, called ME-SOFS, converts applied force into fluid flow that drives actuators, eliminating the need for electronic sensors, signal processors, or external power. The sensor could enable soft robots to operate reliably in wet, hot, or high-pressure environments where conventional electronics fail, opening new possibilities for surgery, deep-sea exploration, and human-machine interaction.

• A new quantum primitive has emerged to tackle one of quantum computing’s biggest bottlenecks. The quantum Hermite transform, implements the Hermite transform on a quantum state, bringing a mathematical operation widely used in physics and machine learning onto quantum hardware. The team designed a quantum circuit that executes the transform with logarithmic overhead rather than linear scaling, offering exponential speedups for large quantum states and allowing quantum computers to directly compute a system’s future state without lengthy simulations. Paired with novel methods for configuring qubits in their correct initial configuration, the algorithm becomes a practical, high-precision primitive capable of analyzing and representing data in new ways. Bao emphasized that quantum computers are powerful but limited without sufficient algorithms, and that new primitives enable solving broader suites of problems relevant to real-world science.

🔊 Industry Insights & Updates

• A 98% task success rate has been achieved by Xiaomi's humanoid robot during testing on a flexible automotive production line, marking a significant step beyond laboratory demonstrations toward real factory deployment. The trial took place at Xiaomi's own automotive manufacturing operations, where the robot completed a round of validation tasks in an environment known for variable fixtures, tight tolerances, mixed-model production, and frequent workflow changes. Xiaomi's dual role as both robot developer and first user gives the company a unique advantage, allowing it to shorten the feedback loop between product development and industrial application by generating operational data from its own car plant. The results position Xiaomi alongside other fast-moving Chinese humanoid robotics players including Unitree Robotics, AgiBot, and Fourier Intelligence, as the sector shifts from staged demonstrations toward real operating data and early-stage commercialization.

• A major leap in robotic spatial perception has been achieved with the launch of LingBot-Depth 2.0 and LingBot-Vision by Robbyant, an embodied AI company within Ant Group. The new spatial perception model, trained on 150 million samples, tops 12 out of 16 depth completion benchmarks and halves depth error in demanding indoor scenarios compared to its predecessor. LingBot-Vision, a novel visual foundation model, is the first in the industry to use boundary structure as a pre-training objective, delivering sub-pixel-level boundary localization with only 160 million training images. Orbbec has professionally certified the technology, and the two companies will collaborate on new SDK products and an integrated camera planned for year-end release. Robbyant has open-sourced the model weights of LingBot-Vision to accelerate commercial deployment of embodied intelligence.

• Mass production of Xpeng's Iron humanoid robot is being ramped up to over 1,000 units per month by the end of 2026, setting the stage for a global roll-out in 2027. The electric vehicle maker plans to deploy the robot as a sales assistant in its China stores during the first quarter of 2027, before expanding to overseas locations later that year. CEO He Xiaopeng is personally leading the robotics business as Xpeng transforms from a smart-car company into what it calls a "physical AI company," repurposing autonomous-driving technologies for humanoid robots. The Iron robot stands 178 centimeters tall, weighs 70 kilograms, runs on all-solid-state batteries, and is powered by three Xpeng-developed Turing AI chips with 2,250 TOPS of computing power. A new production base covering 110,000 square meters is under construction in Guangzhou to support the mass-production plan, with the company expecting to spend RMB 7 billion on physical AI research this year.

• A new world record of 28.04% power conversion efficiency for perovskite-organic tandem solar cells has been achieved by Chinese researchers. Academician Li Yongfang and researcher Meng Lei from the Institute of Chemistry under the Chinese Academy of Sciences led the research, which brings the technology a step closer to practical use. The team overcame long-standing problems of performance degradation and limited stability by introducing a new additive molecule called TDB that regulates the material during both fabrication and operation under illumination. Third-party certification confirmed the record efficiency, and the cell retained 90% of its initial performance after 625 hours of continuous illumination. The technology could eventually be adapted for buildings, transportation, wearable electronics, drones, and satellites where lightweight, flexible solar cells offer advantages over conventional panels.

Subscribe now

🧬 BioTech

• A breakthrough approach to cancer treatment has emerged from researchers at Wayne State University and Indiana University, who have developed drugs that strip cancer cells of their ability to heal themselves after radiation and chemotherapy. Backed by a $3.2 million renewed grant from the National Cancer Institute, Dr. Navnath Gavande of Wayne State and Dr. John Turchi of Indiana University School of Medicine are leading the development of Ku-targeted small molecule inhibitors that target the Ku70/80 DNA-binding complex, a central sensor that cancer cells use to detect broken DNA and recruit repair enzymes. Unlike current approaches that target the DNA-PK enzyme directly and can cause toxicity, the new compounds called Ku-DNA binding inhibitors block the earliest step in the repair process, preventing DNA-PK from being properly activated. During the initial funding phase, the team discovered molecules that enter cells, inhibit DNA-PK activation, and sensitize cancer cells to radiation in preclinical models. The next phase will focus on identifying the cancer types where these inhibitors can have the greatest therapeutic impact and creating new synthetic lethal interactions in difficult-to-treat tumors.

• A key architect of the immune system’s fight against cancer has been discovered by researchers at the Icahn School of Medicine at Mount Sinai. The team identified dendritic cell type 1 as the essential immune cell that builds and maintains tertiary lymphoid structures, organized clusters that act as local command centers where the body coordinates attacks against tumors. The study, published in Science on July 16, 2026, was led by lead author Raphael Mattiuz and senior author Miriam Merad, who analyzed tumor samples from patients with lung, liver, colorectal, kidney, and ovarian cancers. Using a new mouse model, the researchers showed that these dendritic cells do more than alert the immune system to cancer, remaining inside tumors to coordinate both cancer-killing T cells and antibody-producing B cells. The findings could lead to therapies that increase the number or activity of dendritic cells, potentially improving immunotherapy for patients whose cancers do not respond well to current treatments.

• A novel salvage surgical approach has provided durable pain relief for a patient with severe, treatment-resistant multiple sclerosis-related trigeminal neuralgia. The procedure combined internal neurolysis, known as nerve combing, with targeted glycerol delivery to the proximal trigeminal root, achieving immediate pain relief in a 50-year-old woman who had failed medications, two glycerol rhizotomies, and Gamma Knife radiosurgery. The patient remained recurrence-free on the treated side for 43 months with preserved sensation and reduced medication use, according to the case report published in the Chinese Neurosurgical Journal on May 14, 2026.

  ---

💡Products/tools of the week

• Indexing over one million competitor ads and tracking daily performance snapshots is how adWhispr identifies proven winners by how long they've been running. The AI marketing agent lets users name any brand to pull its full ad library, then clone a winning image or video ad with new footage, voiceover, and branding in minutes. adWhispr integrates directly with Claude, ChatGPT, Cursor, and other MCP-compatible apps so research, creative generation, and campaign launch happen within a single conversation. Campaigns are created with budget and targeting set automatically on Meta, TikTok, or Google, and adWhispr always pauses them for user review before any spend occurs. The tool is built for solo founders, marketing agencies, and content creators who want to replace manual ad research and creative production with one chat workflow.

• Connecting AI agents to dozens of external data sources through a single API key and one MCP installation is now possible with AgentKey, a data marketplace and connectivity layer. AgentKey supports agents such as Claude Code, Cursor, Windsurf, Gemini CLI, and Codex, removing the need to manage individual API keys and billing for each service. Supported data categories include web search, web scraping, social media platforms, financial market data, on-chain crypto data, and e-commerce feeds. AgentKey handles failover automatically when a provider goes down, ensuring continuous data access. The platform is designed specifically for AI agent workflows that require real-time, live external data to complete research and analysis tasks.

• Creating, analyzing, and scaling AI-generated video ads in minutes is now possible with Maxfusion AI, a platform built for marketers. It features RIZZ, a proprietary audio-guided video model that generates UGC-style videos with expressive AI actors created from a single face image. Users can upload existing ads for analysis, extract keyframes, and duplicate winning formats quickly. Maxfusion AI supports bulk generation of ad variations with different hooks, scripts, and actors, and can localize ads into 35 or more languages instantly.

• End-to-end testing for iOS, Android, and web applications is now powered by AI through Autosana, a platform that lets engineering teams ship faster with greater confidence. Autosana features self-healing tests that automatically adapt when UI changes occur, using code diffs to update test flows without manual intervention. The platform supports cloud-hosted devices and works with all major frameworks including Flutter, React Native, Swift, Kotlin, and React. Tests can be scheduled on CI/CD triggers or run automatically on pull requests, with Autosana delivering results via session replay, email, Slack, or PR comments.

  ---

• Robbyant: LingBot-VLA 2.0: 20 Robot Configurations, Full-Body Control | Now Open-Source

• 1X: NEO’s Hands

• Booster Robotics: Booster T2 is now available.

• Walden Robotics: Walden Robotics: General-purpose robots that continuously learn & improve while performing real work

Welcome to this week's edition of Rushing Robotics! This week's lineup reads like a sci-fi anthology written by engineers who actually shipped the product. We've got pneumatic gloves that give paralyzed hands their intention back, ultrasonic sensors that make factory floors safe enough for humans and robots to share, and a touch-sensitive material that literally changes color when pressed — no electronics required. This week's lineup is a testament to that compounding effect.

🤯 Mind-Blowing

Where engineering meets the impossible. This section dives into the breakthroughs that make you pause and reread the headline—soft exoskeletons decoding biosignals into movement, chips that compute through memory itself, sensors that see touch as color, and batteries measured in millennia. These aren't incremental updates; they're category redefinitions.

🔊 Industry Insights & Updates

The infrastructure layer is shifting. NVIDIA and Hugging Face are building shared rails for humanoid robotics, MIT is threading light through silicon for petabit communication, Argonne is letting AI agents run chemistry on exascale machines, and Huawei is rewriting the semiconductor roadmap with time-based scaling. The foundations being laid today will determine what's buildable tomorrow.

🧬 BioTech

Biology is becoming programmable. Yeast supplements reprogramming immune cells to fight cancer, humanoid robots assisting in surgery, and small molecule probes cracking previously "undruggable" targets—this section tracks the frontier where medicine, robotics, and molecular engineering converge.

💡 Products/Tools of the Week

The practical layer where breakthroughs become usable. AI benchmarks disguised as Gary Busey portraits, coding agents that can finally see what they're building, privacy policy generators that actually understand local law, and a platform turning single photos into conversational AI agents—these are the tools shipping now that change how we work.

🎥 Video Section

Because some breakthroughs need to be seen in motion. Robot combat competitions, camera bots, startup journeys, and chore-killing humanoids—these are the visuals bringing the week's stories to life.

We're entering an era where paralyzed hands grasp again, where robots assist in surgery and then clean up the room, where a dietary supplement can retrain the immune system, and where the tools to build all of this are increasingly open and shared. The future isn't a distant horizon we're walking toward; it's a wave we're already riding, and it's moving faster than anyone predicted. Stay hungry, stay futurish!

🤯 Mind-Blowing

• Intentional grasping was restored in a patient with severe hand paralysis using a soft pneumatic glove developed by Technical University of Munich researchers. The lightweight textile exoskeleton features wrist dorsiflexion, an active thumb, and 13 air tubes that bend and straighten fingers to enable controlled grasping. A surface EMG grasp predictor reached 97% sensitivity compared with healthy controls, using motion data and machine-learning correction to decode noisy biosignals into reliable hand movements. The team reported that follow-on validation included stroke patients, with the most encouraging results seen in cases of severe or near-complete impairment, while utility for moderate residual function appeared more task dependent. The work highlights how noisy biosignal decoding becomes useful only when paired with safe, wearable actuation evaluated against the impairment level it is meant to support.

• A world-first safety certification for 3D ultrasonic sensing was unveiled by Sonair, a Norwegian robotics technology company, with the launch of its ADAR One sensor. The device became the first 3D sensor certified to SIL 2 and PL d safety standards using sound instead of light, meeting the European machine directive for detecting humans and objects around robots. CEO Knut Sandven described the milestone as solving the safe perception bottleneck that has limited human-robot collaboration. The sensor delivers 180°×180° spatial awareness, eliminating the blind spots that plague traditional 2D laser scanners, and is already shipping on deployed industrial robots after evaluation by more than 80 global robotics companies.

• Robotic touch took a colorful turn as scientists built a new sensor material that shifts hue when pressed, eliminating the need for complex electronics. A simple USB camera watches the color patterns and creates an instant map of where and how strongly an object makes contact, skipping the heavy data processing required by traditional touch sensors. The study, published in Science Advances, brought together researchers from the United Kingdom and Italy to merge advances in soft robotics and flexible materials. Giacomo Sasso, who led the research, highlighted that the sensor can capture details as fine as fingerprint ridges, a level of precision no existing technology matches at such low complexity. Applications could span delicate factory assembly, robotic surgery that distinguishes healthy from diseased tissue, and prosthetic hands that let users handle fragile objects safely.

• The first neural dynamical system chip built on phase-change memristor technology was unveiled by a research team at Peking University led by Professor Yang Yuchao. The chip achieves a single-step computation latency of just 2.12 milliseconds, bringing neural dynamics hardware into the millisecond era for the first time. The breakthrough uses a "controlled in-memory computing" paradigm that lets the physical evolution of memory materials perform computations, eliminating the data movement bottleneck of traditional architectures. For brain cortex surface reconstruction tasks, the chip achieves up to 478x acceleration over NVIDIA A100 GPUs, with major implications for brain-computer interfaces, digital brain twins, and early screening for neurodegenerative diseases such as Alzheimer's and Parkinson's.

• A new-generation nuclear battery with a theoretical lifespan of thousands of years was unveiled by Northwest Normal University and Gansu Zhulong Technology, marking a major advance in long-life power sources. Called Qianjiyuan Tianshu, the carbon-14 battery was developed entirely without foreign technology or parts and delivers a 15.5-fold increase in volumetric power density over the team's earlier prototype. Su Maogen, who leads the project team at Northwest Normal University, said the device works reliably between minus 100 and 200 degrees Celsius, making it suitable for medical implants, deep-sea missions, polar regions, defence and aerospace applications. The battery measures just 16.8 cubic cm and uses only 22 per cent of the radioactive material required by its predecessor while boosting short-circuit current 2.5 times and maximum power 2.6 times.

• A fully autonomous robotic pharmacy kiosk that fills prescriptions in 60 seconds was launched by startup Queue in a Palo Alto pilot site, emerging from stealth with $12.6 million in seed funding. The machine takes sealed wholesale pill bottles and produces verified, labeled prescription vials with no on-site pharmacy staff involved, using computer vision to identify every pill at the National Drug Code level. Queue claims the system can fill a 60-pill vial every 30 seconds, reaching roughly 600 pills per minute at full configuration, and currently supports around 250 commonly prescribed medications. Backed by $18.6 million in total funding, the company plans a broader commercial rollout across retail pharmacies, hospitals, clinics, and rural communities by early 2027.

🔊 Industry Insights & Updates

• A collaboration between NVIDIA and Hugging Face will bring the NVIDIA Isaac GR00T 1.7 robot foundation model and the Isaac Teleop data collection framework into LeRobot, Hugging Face's open source robotics library. The move gives developers a standardized path to train, evaluate, and deploy humanoid robot models using shared data and workflows. NVIDIA Cosmos 3, a frontier world foundation model for physical AI, is also planned to join LeRobot soon, helping developers generate synthetic robotics data when real-world data is scarce. Thomas Wolf, cofounder and chief science officer at Hugging Face, said open source lets developers study, adapt, and build on advanced research together.

• A major step toward faster, greener data communications was achieved through MIT's FUTUR-IC research program, which developed new devices that integrate electronics with photonics on microchips. Led by Anu Agarwal, the team created the first "optical bumps" — couplers that connect photonic chips to electronic ones, similar to how metal solder bumps connect traditional chips. Two key devices, the evanescent coupler and the graded index coupler (GRIN), were reported in Advanced Engineering Materials and the Journal of Physics: Photonics. The work promises to push data transmission from hundreds of terabits per second to over 1 petabit per second while cutting energy use, since light-based communication is far more efficient than electricity. Funded by the NSF Convergence Accelerator, the program also developed Earthster, a sustainability tracking tool for companies, and launched workforce training programs including online courses and bootcamps.

• A breakthrough in computational chemistry arrived as Argonne National Laboratory researchers unveiled ChemGraph, an open-source framework that uses AI agents to automate complex simulations. Built on the Aurora exascale supercomputer, the system lets researchers describe scientific problems in plain language instead of manually navigating technical workflows. The framework combines large language models with agent-based automation, splitting tasks between larger models for planning and smaller ones for execution to reduce costs. Thang Duc Pham and Murat Keçeli led the project, which could advance battery design, combustion systems, and materials discovery. The research was published in Communications Chemistry.

• A new semiconductor principle called the Tau (τ) Scaling Law was unveiled by He Tingbo from Huawei at the 2026 IEEE International Symposium on Circuits and Systems in Shanghai on May 25, 2026. The law replaces traditional geometric scaling with time scaling as the guiding principle for semiconductor evolution, using technologies like LogicFolding to compress signal propagation delay and improve transistor density. Huawei has already designed and mass-produced 381 chips based on this principle over the past six years, with Kirin chips launching in Fall 2026 as the first to adopt the LogicFolding architecture.

Subscribe now

🧬 BioTech

• A yeast-based food supplement was shown to help the body make stronger cancer-fighting immune cells, in a first-of-its-kind study by researchers from Trinity College Dublin and University College Dublin. Working in a mouse model, the team led by Frederick Sheedy and Helen Roche found that adding yeast beta-glucan to the food of obese laboratory mice changed how their immune cells grow, producing better anti-tumour cells. The findings, published in Cell Reports, show that the dietary intervention restores anti-tumour immunity in obese mice and reverses long-term immune memory defects that persist even after weight loss. First author Anna Ledwith said the study tested whether the supplement could reprogramme early-stage immune cells in bone marrow to produce enhanced immune responses against colorectal, skin and breast cancer.

• Two successful surgeries performed with teleoperated humanoid robots marked a medical breakthrough by researchers. The team at UC San Diego’s Center for the Future of Surgery ran a preclinical trial on large non-primate mammals in which a humanoid robot named Surgie removed a gallbladder while a human surgeon assisted, followed by a fully robot-robot procedure with two humanoid robots operating together. Shanglei Liu, MD, teleoperated Surgie and said the precision matches mature robotic systems while needing far less cost and physical space. Unlike conventional surgical robots that are large and task-specific, Surgie weighs 60 pounds, stands five feet tall, and can function in cramped or under-resourced settings without expensive room modifications. The humanoid design also lets the robots fetch tools and handle post-operative cleanup, positioning them as full surgical team members rather than isolated devices.

• A small molecule probe called PBITE-1 was developed by University of Michigan researchers to target a protein called ERG, which was previously considered undruggable in prostate cancer. The study, published in Proceedings of the National Academy of Sciences, showed that PBITE-1 can bind to a specific pocket within ERG and disrupt its ability to drive tumor development. Led by Arul Chinnaiyan, M.D., Ph.D., the team screened over 1,600 compounds and found that PBITE-1 causes cancer cell death and prevents invasion of nearby tissues in preclinical models. About half of prostate cancer cases in patients of European ancestry involve a gene fusion that turns on ERG, making it a key driver of tumor growth. The findings open a path toward personalized therapy for prostate cancer subtypes.

  ---

💡Products/tools of the week

• A new benchmark called BuseyBench tests over 230 language models by having them generate SVG portraits of actor Gary Busey. The benchmark evaluates models from providers like OpenAI, Anthropic, Google, and DeepSeek, scoring each attempt on output quality, token usage, cost, and duration. Its leaderboard lets users filter and sort by provider, model family, score, and cost. Despite the absurd premise, BuseyBench offers real insight into how AI models handle structured creative generation tasks like SVG image construction.

• A new tool called AgentScreenshots lets coding agents like Claude Code and Codex capture precise UI screenshots on demand. Using a single CLI command, it delivers headless rendering in about 2 seconds for localhost or live sites. Agents can target specific components with CSS selectors, simulate clicks, hover, and scroll, pick viewports, and use supplied instruction packages. This lets LLM agents see the UI to verify edits, run visual checks, and iterate faster without noisy navigation logs or bloating model context.

• An AI-powered privacy and compliance document generator called PolicifyAI scans your site for GDPR and CCPA gaps and produces tailored, jurisdiction-specific policies in under 60 seconds. It can generate 120-plus document types in 120 languages for 180 jurisdictions, translating legal terms rather than just body text, and exporting to PDF, DOCX, or live-embedded HTML. The AI drafts clauses based on actual local laws, quality-scores each output for completeness and clause coverage, auto-regenerates drafts that fall short, and keeps embedded policies up to date so businesses can quickly get legally-relevant templates without hiring a lawyer.

• Ojin enables users to create human-like AI agents from a single photo. Developed by Journee Technologies, it uses advanced speech-to-video models to power fully conversational agents capable of real-time dialogue, emotional nuance, and natural presence. The platform offers multiple tiers including the Human Agent for lifelike conversational interactions and API-accessible models like Oris Portrait and Oris Presence for scalable or high-fidelity deployments. Ojin is suited for solo creators through to enterprise teams needing custom volume, SLAs, and dedicated support.

  ---

• Unitree Robotics: Unitree Invites You to Witness the 2026 Humanoid Robot Combat Competition

• Mondo Robotics: Meet Beni, Your First Camera Robot by Mondo Robotics

• Austen Hartley: i started a robotics company in 2026

• OTOFOOTAGE: This $8000 Robot Just Killed House Chores Forever – Weave Isaac 1 Full Demo

Welcome to this week's edition of Rushing Robotics, where the boundaries between science fiction and reality continue to blur at an astonishing pace. From synthetic cells completing full life cycles to sub-nanometer chips packing 100 billion transistors, from humanoid robots entering classrooms to diamond semiconductors reshaping the future of computing—this week's breakthroughs span the entire frontier of human innovation. We're witnessing a convergence of biology, robotics, materials science, and artificial intelligence that is fundamentally rewriting what's possible. Whether it's fungi being engineered to produce never-before-seen anti-cancer compounds or AI-powered tools transforming how we write code, design spaces, and generate fonts, the pace of progress has never been more relentless—or more exciting.

🤯 Mind-Blowing

This week’s discoveries push the limits of what we thought possible: a synthetic cell built entirely from non-living chemicals completing a full life cycle, the first AI-powered humanoid robot deployed as a teacher’s assistant in a U.S. public school, IBM’s groundbreaking 0.7-nanometer chip technology, a laser technique that listens to the quantum forces inside proteins, and China’s launch of a major diamond semiconductor manufacturing hub. Each story represents a leap forward that would have seemed impossible just years ago.

🔊 Industry Insights & Updates

The robotics and energy sectors are accelerating at full throttle. Apptronik unveiled a massive 90,000-square-foot Robot Park in Austin alongside Apollo 2, researchers in Germany set a new 25.5% efficiency record for CIGS-perovskite tandem solar cells, BMW confirmed deployment of the next-generation Figure 03 humanoid robot at its Spartanburg plant, and Agility Robotics announced plans to go public through a $2.5 billion merger—making it the first U.S. humanoid robotics company to enter public markets.

🧬 BioTech

The frontier of biological engineering is expanding rapidly. A new genome editing tool called fPE7max has unlocked eight never-before-seen molecules from fungi, three showing anti-cancer activity. Children's National reported encouraging results from a first-in-human T-cell immunotherapy trial for pediatric brain tumors. And researchers revealed how bacterial enzymes collaborate to create multiple versions of anti-cancer drugs, potentially accelerating development of new treatments.

💡 Products/Tools of the Week

This week's launches include Prbl's AI-first code security scanner for vulnerabilities introduced by AI coding tools, FontDance's AI-powered font generator, Hausey's interior design and virtual staging platform, and FluidVoice's free, open-source macOS voice-to-text app with on-device AI polish.

🎥 Video Section

This week's video highlights feature Flexion Robotics' Reflect V1.0 on long-horizon autonomous humanoid work, Pudu Robotics' collaboration with Shenzhen CTID Co. Ltd, and AGIBOT's livestream of humanoid robots on a real production line.

AI is no longer just generating text and images; it's designing therapies, discovering materials, orchestrating robots, and reshaping energy systems. Each of these stories would have been a decade-defining headline on its own. Together, they suggest we're entering a phase where the bottleneck isn't imagination or capability, but simply how fast humanity can responsibly deploy what's now possible. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A synthetic cell built entirely from non-living chemicals has become the first in the world to complete a full life cycle, marking a historic breakthrough in biological engineering. Associate Professors Kate Adamala and Aaron Engelhart, along with their teams at the University of Minnesota College of Biological Sciences, created the cell, which they named SpudCell. The synthetic organism can grow, replicate its genome, feed on resources, and divide, all without any living components. SpudCell achieves division without a cytoskeleton, using proteins that crowd together on the membrane surface until mechanical stress splits it in two. Researchers even observed natural selection in action, as a faster-growing variant outcompeted the original after just five generations. With a genome of only 90 kilobase pairs spread across seven plasmids, SpudCell is smaller than biologists thought possible for a living cell. Adamala and her partners are now launching Biotic, a public-benefit research institution, to build shared infrastructure for synthetic cell engineering worldwide.

• A pilot program deploying an AI-powered humanoid robot as a teacher's assistant was unanimously approved by the Salamanca City Central School District board on June 6, marking the first such deployment in a U.S. public school. The district, approved spending $57,590, with $50,000 going to the robot hardware and the remainder covering subscription services and shipping. Las Vegas-based Realbotix Corp. is supplying its M-Series humanoid, a 39-degree-of-freedom robot with a modular upper body and micromotors in its face that power lifelike expressions. The robot normally starts at $95,000, so the district received a substantial discount in exchange for giving Realbotix bragging rights for the historic deployment. Teachers and staff will use a controlled interface to upload district-approved lessons, and the robot will not connect directly to the internet or crawl internal servers. Realbotix CEO Andrew Kiguel called it a landmark moment for AI and humanoid robotics. Superintendent Dr. Mark Beehler said the system will give students a safe, Salamanca-specific AI tutor and educators customized AI tools. The pilot will initially support high school students in Woz ED AI and robotics courses, part of Steve Wozniak's STEM pathways program, with plans to expand to 500 students in Fall 2026.

• A sub-1 nanometer chip technology has been unveiled by IBM, marking the world's first breakthrough at the 0.7 nanometer, or 7 angstrom, node. The new chip packs nearly 100 billion transistors onto a fingernail-sized surface, nearly twice the density of IBM's 2 nanometer chip introduced in 2021. IBM developed a revolutionary transistor architecture called "nanostack," the industry's first three-dimensional, nanosheet-based design, which vertically stacks and staggers transistors to pack more of them onto a single chip. The technology is projected to deliver up to 50 percent more performance or 70 percent greater energy efficiency than IBM's 2 nanometer chips, boosting applications from generative AI to cloud infrastructure. Jay Gambetta, Director of IBM Research and IBM Fellow, said the innovation pushes technology beyond the nanometer era to the scale of atoms and reinvents how chips are built. IBM's semiconductor roadmap projects at least a decade of future scaling, with production possible in as early as five years.

• A laser technique that directly measures quantum forces in proteins has been invented by researchers at Texas potentially transforming how drugs are designed and tested. The method, called Thermostable Raman Interaction Profiling or TRIP, fires a laser at a sample and records the vibrational signals that bounce back, letting scientists listen to the internal melody of molecules. Dr. Narangerel Altangerel, lead researcher and assistant research scientist, said the team can now directly measure molecular forces at their most fundamental level. The researchers tested TRIP on the main protease of SARS-CoV-2, the virus behind COVID-19, and found it could accurately predict how well antiviral drugs would bind and work. Dr. Philip Hemmer, professor of electrical and computer engineering and a collaborator on the study, said the technique is non-invasive and could speed up drug prescreening. Altangerel and her team have filed a U.S. patent for the invention.

• A major diamond semiconductor manufacturing hub has been launched in Zhengzhou, marking China’s first fully integrated domestic ultra wide bandgap semiconductor project. The project will cover the entire production process, including equipment manufacturing, single-crystal diamond growth, epitaxial processing, and semiconductor substrate packaging, using Sino Powder’s proprietary micro- and nanodiamond production technology. The production base will house 500 microwave plasma chemical vapor deposition systems capable of producing 2-inch and 4-inch single-crystal diamond wafers, along with 50 production lines for spherical diamond powder. The first phase will bring 200 MPCVD systems online by the end of 2026, with full operations expected to generate annual output worth around 3 billion yuan within three years. Ultra wide bandgap materials like diamond are considered promising for future semiconductor devices due to their superior thermal conductivity, high breakdown voltage, and ability to operate under extreme conditions, making them suitable for AI processors, electric vehicle power electronics, 5G and 6G communications, and advanced medical equipment.

🔊 Industry Insights & Updates

• A newly expanded Robot Park facility for humanoid robot training has opened in Austin, Texas, marking a major step forward in the development of AI-powered robotics. Apptronik, the company behind the project, unveiled the nearly 90,000-square-foot flagship data collection and training center on June 30, 2026. The facility anchors a growing global network of Robot Parks at customer and partner sites worldwide. Alongside the opening, Apptronik introduced Apollo 2, the latest version of its humanoid robot platform, available in both bipedal and wheeled-base configurations. Apollo 2 robots continuously gather real-world data across logistics, manufacturing, retail, and other customer-driven activities. This data fuels the development of Gemini Robotics, Google DeepMind's foundational AI models for robotics, through an ongoing research partnership. Apptronik CEO and co-founder Jeff Cardenas said the company is focused on what robots can do every day on the job, not just in demos. Fleets of Apollo 2 robots are already active at Robot Park locations including Google DeepMind, Mercedes-Benz, and GXO sites.

• A new record efficiency of 25.5% has been achieved for a CIGS-perovskite tandem solar cell by researchers in Germany. The work was part of the EU-funded SOLMATES project, led by HZB to explore combining CIGS and perovskite technologies, and surpasses the group's previous record of 24.6%. HZB researcher Guillermo Farias Basulto said the team used CIGSe-bottom cells with different band gaps of 1.05eV and 1.1eV and two different thicknesses of aluminium-doped zinc oxides. To reduce interfacial recombination losses and improve stability, the researchers screened combinations of nickel oxide and self-assembled monolayers as hole transport layers. They also refined electron-selective contact processing by regulating the thermal evaporation rate of Buckminsterfullerene onto a 1nm lithium fluoride passivation layer. The record cell covered only 1.081cm², but the team also fabricated a mini-module reaching 19.7% efficiency across 2.25cm². Farias Basulto said the physics of the current architecture suggests 25.5% is merely a stepping stone, with in-house testing of similar designs already reaching 27.5%.

• Deployment of the next-generation Figure 03 humanoid robot has been confirmed at BMW Group Plant Spartanburg, where it will handle logistics sequencing in a real production environment. The announcement follows a ten-month pilot in which the Figure 02 robot, built by Figure AI and its founder and CEO Brett Adcock, inserted sheet-metal parts for welding on more than 30,000 BMW X3 vehicles. BMW Group said the new robot introduces soft components for enhanced safety, wireless charging for higher availability, audio functions for speech-to-speech communication, and improved hands with tactile sensors and palm cameras for greater precision. Figure 03 will pick unsorted components from large containers and sort them into a sequencing trolley, which will then be moved by an automated tugger train or Smart Transport Robot to the assembly line for just-in-sequence delivery. Ulrich Wieland, Vice President of Production Control and Logistics at BMW Manufacturing, called Spartanburg the birthplace of humanoid robotics in BMW's operational day-to-day activities and said the company looks forward to deploying Figure 03 for a sequencing use case in logistics.

• A historic step for humanoid robotics was taken as Agility Robotics revealed plans to go public through a $2.5 billion merger with special purpose acquisition company Churchill Capital Corp XI, making it the first U.S. company focused exclusively on humanoid robots to enter the public markets. Agility's robot, Digit, is a two-legged machine about 5-foot-9 tall, designed to work in facilities built for humans, lifting up to 35 pounds and handling repetitive tasks like moving totes and tending machines for up to 20 hours a day. The forthcoming Digit v5 will add swappable hands, a 50-pound lifting capacity, and safety systems intended to let it operate alongside people without barriers. The company also became the first to adopt NVIDIA's newly announced Halos safety system for robotics, part of a broader push to deploy humanoid robots that can work safely alongside humans.

Subscribe now

🧬 BioTech

• A new genome editing tool called fPE7max has been developed to unlock the hidden drug-producing potential of fungi, yielding eight molecules never before seen by science, three of them showing early anti-cancer activity. Xue "Sherry" Gao, a chemical and biomolecular engineer at the University of Pennsylvania's School of Engineering and Applied Science, led the team that created the platform to navigate the complex genetics of filamentous fungi like Aspergillus and Penicillium. First author Chunxiao Sun, a postdoctoral researcher in the Gao Lab, said one novel molecule showed selective toxicity against human breast, hepatic, and leukemia cancer cells. The findings were published in Nature Biotechnology. Fungi have long been the forgotten middle child of genomics, turning off their drug-producing gene pathways when grown in sterile lab conditions. Gao's team overcame this by adapting prime editing, a technology that avoids double-strand breaks, and integrating a protective protein called fLa to shield fragile RNA instructions during massive DNA insertions. The platform achieves editing efficiency approaching 90 percent.

• Encouraging early results from a first-in-human clinical trial using a new T-cell immunotherapy for children and young adults with deadly brain tumors have been reported by researchers at Children's National. The Phase 1 study, published in Nature Medicine, evaluated a multi-targeted T-cell therapy designed to strike three proteins commonly found in pediatric brain tumors: WT1, PRAME, and Survivin. Catherine Bollard, senior vice president and chief research officer at Children's National and co-senior author of the study, said the research represents an important step toward developing safer and more effective T-cell therapies for children with devastating brain cancers. Eugene Hwang, chief of Oncology at Children's National and co-senior author, said the team was excited to preserve safety and quality of life while generating anti-tumor responses by attacking three targets at once. The therapy is delivered through the bloodstream rather than direct injection into the brain, producing fewer severe side effects than some existing engineered immune therapies. Some patients remained disease-free years after treatment.

• A breakthrough in understanding how bacteria create multiple versions of anti-cancer drugs has been achieved potentially accelerating development of new treatments for hard-to-treat cancers. The researchers revealed how bacterial enzymes communicate and work together to assemble a family of related anti-cancer compounds, including Romidepsin (Istodax), a clinically approved blood cancer treatment. Published in Nature Communications, the study shows that small molecular regions called docking domains act as connectors between the core drug assembly machinery and the variable component-building enzymes. These docking domains use a conserved connection point compatible with multiple different enzyme partners, which explains how bacteria generate structural diversity while keeping their drugs precise and effective. Dr. Munro Passmore, Research Fellow in the Department of Chemistry at the University of Warwick and first author, said the system is so elegantly economical that it had eluded researchers for decades. Prof. Greg Challis, Monash Warwick Alliance Professor of Sustainable Chemistry at both universities, said the discovery moves the field from understanding how the systems work to building new ones.

  ---

💡Products/tools of the week

• A new AI-first code security scanner launched to address vulnerabilities introduced by AI coding tools. Prbl detects which files and lines were generated by models and scans specifically for classes of vulnerabilities that AI coding tools commonly introduce, including hardcoded secrets, injection flaws, missing authentication, fallback secrets, and timing-unsafe checks. The company built an automated rewriter that proposes minimal, behavior-preserving fixes validated against a baseline and a rescan. Development teams can use Prbl alongside their existing SAST tools to find and safely remediate issues that general scanners miss in AI-generated code. The product ships with CI and GitHub integration plus PR comments to streamline fixes.

• FontDance released an AI-powered font generator that transforms a style prompt or reference image into typeable, exportable TTF and WOFF2 fonts in minutes. The generator produces consistent glyphs, checks structure, spacing, and readability, and provides a live text preview so users can test rhythm before downloading. FontDance built the tool so teams can rapidly prototype usable web and brand fonts, explore creative lettering directions without manual glyph drawing, and get production-ready files with originality guardrails and quality-aware AI checks. The company positioned the product as a faster alternative to traditional font design workflows that require manual glyph drawing and extensive kerning work.

• Hausey released an AI-powered interior design and virtual staging tool that turns a single photo of a room or exterior into ten photoreal, designer-grade remodel directions in under 60 seconds. The tool uses generative models plus automated structural-consistency checks to produce MLS-safer, refinable visuals that users can share, collect votes on, and export as HD images or PDFs. Hausey built the platform for contractors, design-build firms, real estate agents, designers, and homeowners who need fast, believable before/after presentations to speed decisions and close deals. The company positioned the product as a way to replace slow, costly rendering workflows with instant, shareable design options.

• A free, open-source macOS voice-to-text app launched to give users fast, offline transcription with on-device AI polish. FluidVoice uses local speech models for transcription and an optional on-device AI layer called Fluid-1 or Fluid Intelligence to post-process and polish transcripts, fixing punctuation, capitalization, dates, numbers, structure, and adapting tone per app. The company built the app so dictation types into any app as clean, context-appropriate text with zero data leaving the Mac by default. FluidVoice also lets users opt to use cloud AI providers such as OpenAI, Groq, or custom options if they need additional enhancements. The company said the product is ideal for privacy-conscious users, heavy typists, and anyone who wants faster, more natural writing on macOS.

  ---

• Flexion Robotics: Reflect V1.0 - The Path Towards Long-Horizon Autonomous Humanoid Work

• Pudu Robotics: Pudu Robotics × Shenzhen CTID Co. Ltd

• AGIBOT: Humanoid Robots on a Real Production Line: AGIBOT Livestream

Welcome to this week's edition of Rushing Robotics! While the world argues about what AI might do someday, the labs, startups, and factories are quietly shipping things that would have looked like magic just a few years ago. This week's lineup is no exception — we've got self-healing lenses, mile-deep nuclear reactors, brain chips thinner than a hair, and the world's first HIV-to-HIV lung transplant.

🤯 Mind-Blowing

We open with the kind of stories that make you pause and reread. Self-healing contact lenses that fix their own scratches under a UV lamp. AI-nominated CAR T cell targets that took weeks instead of years to identify. A flexible, all-organic brain chip that held 94% signal strength after 550 days in vivo. Plus, wave-powered floating data centers and a 18.5 GW pipeline for mile-deep nuclear reactors. This is the section where the impossible starts looking routine.

🔊 Industry Insights & Updates

Then we move to the technologies crossing from lab to market. Boron nitride nanotube membranes enabling ultrafast ion transport. Sodium-ion batteries completing 15,000 km in heavy-duty trucks at -40°C. CATL's humanoid robots working their own production lines. And Google partnering with Energy Dome to bring CO2 Battery long-duration storage to Ireland. These are the signals of where capital, engineering, and policy are aligning.

🧬 BioTech

We turn to the life sciences, where the stakes are personal. Inhaled vitamin D as a rethink for COPD therapy. Phytochemicals modulating miR-29a for cancer prevention. And a world-first: an HIV-positive donor lung transplant at NYU Langone that gave a patient back his breath after four years on oxygen. Medicine is quietly rewriting its rulebooks.

💡 Products/Tools of the Week

A quick detour into practical releases: CanIRun.ai for local AI model compatibility checks, AllyHub for agent-driven research workflows, RowSpeak for natural-language spreadsheet analysis, and Rendervi for fast architectural rendering.

🎥 Video Section

Finally, we close with movement: DEEP Robotics' Lynx M20S on all terrain, ANYbotics inspecting a 150-year-old cement plant, AGIBOT's G2 humanoids on a real tablet line, and MindOn's single-model, multi-robot logistics run. Watch the machines learn to work.

AI is no longer just generating text and images; it's designing therapies, discovering materials, orchestrating robots, and reshaping energy systems. Each of these stories would have been a decade-defining headline on its own. Together, they suggest we're entering a phase where the bottleneck isn't imagination or capability, but simply how fast humanity can responsibly deploy what's now possible. Stay hungry, stay futurish!

🤯 Mind-Blowing

• Self-healing contact lenses that repair scratches with UV light have been developed by researchers Jung-Hyun Choi and Byoung-Ki Cho. The hydrogel material uses a disulfide cross-linker and methacrylate polymer that undergoes disulfide exchange when exposed to UV light at 365 nanometers for one hour, allowing sulfur bonds to break and reform to heal damage. The researchers applied an anti-scratch and antibacterial coating that reduced transparency by only about 2% after sandpaper abrasion testing. The team had previously created a hydrogel that required heating, but extended heat caused lenses to dry out, prompting the switch to UV light at room temperature. Cho noted that repairs can be repeated and performed with at-home UV lamps. The National Research Foundation of Korea funded the research.

• An AI-driven approach to finding CAR T cell therapy targets has been developed by researchers at Penn's Perelman School of Medicine and Abramson Cancer Center. The team built a CAR T targeting GPNMB, the top AI-nominated candidate, which showed robust tumor-killing activity in mouse models of melanoma, leukemia, and colorectal cancer. Lead author Daniel Baker worked under the mentorship of Carl June and Zoltan Arany, combining four public skin cancer datasets with frontier large language models and repeating simulations 1,000 times to reduce hallucinations. The process took weeks rather than months or years. June said this represents one of the first uses of LLMs in cell and gene therapy, and the framework was designed to be modular and disease-agnostic.

• A new flexible brain chip developed by researchers has maintained 94% of its signal strength after 550 days of testing in rabbits. The all-organic hydrogel electrode array, called CHIP, eliminates metal entirely and achieved a conductivity of approximately 2,512 S/cm while measuring just 9 micrometers thick, thinner than a human hair. The 128-channel array demonstrated minimal inflammation and little glial scarring during the 18-month trial, addressing a longstanding problem where conventional rigid metal implants lose performance within one to three years. The manufacturing process involves pre-anchoring the hydrogel to an ultrathin parylene substrate and patterning electrodes via photolithography while the material is dry, solving the shape-instability problem that previously made hydrogels unsuitable for high-density electrode arrays. While the results are striking, researchers caution that rabbit cortex differs substantially from human cortex, and the path forward requires non-human primate studies, toxicology testing, and regulatory review before any human trials could begin.

• A floating data center platform powered entirely by ocean waves has been introduced to the market by Panthalassa, a Washington-based start-up, with the unveiling of its Ocean-3 concept designed specifically for modular AI computing operations. The autonomous platform generates electricity as wave motion forces water through internal turbines, eliminating the need for anchors, seabed cables, or land-based power connections while relying on satellite transmission for data communication. Company leadership estimates energy costs could reach as low as $0.02 per kWh, making it potentially the most affordable energy source available. The company plans to deploy multiple Ocean-3 units that work cooperatively as a single distributed data center, though challenges remain regarding durability in harsh marine conditions and satellite bandwidth limitations for high-transfer workloads.

• A customer pipeline representing up to 18.5 gigawatts of potential generation capacity has been announced by Deep Fission. The Berkeley, California-based advanced nuclear energy company is developing small modular pressurized water reactors installed one mile underground, with CEO and co-founder Liz Muller stating that the growing pipeline reflects urgent interest in the company's mile-deep deployment model. Deep Fission is currently advancing its first reactor project at the Great Plains Industrial Park in Parsons, Kansas as part of the Department of Energy's Reactor Pilot Program, having completed drilling of its first data acquisition borehole to approximately 6,000 feet deep. The company plans to apply for a commercial license with the U.S. Nuclear Regulatory Commission in the first half of 2027, targeting first commercial operations in 2027-2028. The LOIs are non-binding and do not contain commitments to purchase electricity, finance projects, construct facilities, grant exclusivity, or deploy a specified number of reactors.

🔊 Industry Insights & Updates

• A breakthrough in nanoscale ion transport has emerged from the University of Illinois Chicago, where researchers developed boron nitride nanotube membranes that enable ultrafast and highly selective ion movement. The team, led by associate professor of chemical engineering Sangil Kim, created membranes using millions of aligned boron nitride nanotubes as transport channels. When tested with solutions of varying salinity, these membranes showed ion transport rates far exceeding theoretical predictions. The researchers demonstrated practical application by powering simple devices including a watch and calculator using electrical output generated from salt concentration differences, showcasing blue energy system potential. Applications could extend to lithium recovery from spent batteries, energy generation from salinity gradients, desalination systems, and next-generation separation technologies. The team continues investigating the mechanism behind the unusually high transport speeds while scaling the system for industrial use.

• Humanoid robots powered by CATL batteries have begun working on the company's smart production line after CATL signed a strategic cooperation deal with Galbot to scale embodied-intelligence robots across its factories. The Galbot S1, with 50-kilogram dual-arm payload capacity and vision-only centimeter-level positioning, handles material picking and transport in module and battery pack production. CATL's cells use particle-grading cathode technology and a bionic self-healing electrolyte to deliver 8-hour battery life with parts-per-billion failure rates. The partners will also extend CATL's Ning Service after-sales system to cover the robots and establish the world's first aftermarket standard for embodied intelligence. Galbot, founded in 2023 by Peking University researcher Wang He, is the only embodied-intelligence company CATL has invested in.

• A successful 15,000-kilometer test of sodium-ion batteries in heavy-duty trucks has been completed by FAW Jiefang and Zhongke Haina, demonstrating strong performance under extreme cold and fast-charging conditions. The 339 kWh battery pack retained over 90% of usable capacity at -40°C and achieved full charges in 20 to 25 minutes, with a cycle life surpassing 8,000 cycles under fast charging. Zhongke Haina said the battery's thermal stability and milder chemistry match the operational needs of commercial vehicles facing high-frequency heavy loads and all-weather conditions. The results place FAW Jiefang in the competitive sodium battery race alongside CATL, which is building a 40 GWh plant in Fujian, and BYD, which claims a sodium battery capable of more than 10,000 charging cycles.

• A new 23MW / 200MWh long-duration energy storage plant will be built at a former thermal power station near Rhode, County Offaly, as part of a partnership between Google and Energy Dome announced on June 23, 2026. The project uses Energy Dome's CO2 Battery technology to absorb surplus renewable energy during periods of oversupply and dispatch firm power during periods of system stress, enabling smart and cost-effective utilization of grid resources. Google and Energy Dome aim to establish a blueprint for how long-duration energy storage technology can contribute to a clean and affordable electricity system in Ireland. The initiative advances Google's ambition to expand access to 24/7 clean, affordable energy for grids globally and follows a long-term agreement the companies announced last year to deploy the technology at scale.

Subscribe now

🧬 BioTech

• Inhaled vitamin D supplementation rather than oral delivery should be explored in human clinical trials as a potential treatment for chronic lung diseases. COPD affects more than 30 million Americans and is the fourth leading cause of death worldwide, with patients typically having low vitamin D levels that lead to increased exacerbations and worse lung function. Oral vitamin D supplements have not improved lung outcomes because the vitamin is potentially inactivated by an enzyme in the lung's blood vessels. Research using cell and animal models suggests inhaled or nebulized vitamin D can protect against dust, pollution, and pathogens. The authors stated that future research should focus on evaluating inhaled vitamin D therapies in human clinical trials to determine whether direct lung delivery would deliver therapeutic benefits.

• Natural bioactive compounds may offer a promising strategy for cancer prevention by modulating microRNA-29a expression, according to a study published in Current Molecular Pharmacology. The review analyzes miR-29a as a pivotal regulator in cancer development, functioning either as a tumor suppressor or oncogene depending on cellular context, influencing metastasis, proliferation, and apoptosis. Alhujaily outlines how phytochemicals including curcumin, EGCG, resveratrol, genistein, and quercetin restore normal miR-29a expression through epigenetic remodeling and inhibition of oncogenic signaling pathways such as TGF-β/Smad, NF-κB, and PI3K/Akt. These compounds demonstrated anti-cancer effects in preclinical studies by targeting downstream effectors like DNMTs, MCL-1, and CDK6. Alhujaily emphasized that most evidence remains in vitro and challenges including bioavailability and target specificity must be addressed before clinical translation, advocating for comprehensive studies employing multi-omics and systems biology approaches.

• A groundbreaking lung transplant from an HIV-positive donor to an HIV-positive recipient was successfully performed at NYU Langone Health on March 21, 2026, marking the world’s first such procedure. The surgery was carried out under the HIV Organ Policy Equity (HOPE) Act research framework, with Dr. Sapna Mehta, clinical director of the NYU Langone Transplant Institute, leading the effort alongside Dr. Mark Sonnick, transplant pulmonologist and co-author of the research protocol. The recipient was 56-year-old Bertrand Nelson, who had been living with HIV for nearly 26 years and was suffering from severe lung and liver disease. Nelson underwent a dual-organ transplant, receiving both new lungs and a new liver during the same operation. Following the procedure, he no longer requires supplemental oxygen for the first time in four years. The landmark surgery demonstrates that HIV-positive donor lungs may become a viable option for carefully selected HIV-positive transplant candidates, potentially expanding access to lifesaving organs for people living with HIV.

  ---

💡Products/tools of the week

• A new browser-based hardware detection tool called CanIRun.ai launched to help users determine which AI models their local machines can run. The site uses WebGPU browser APIs to instantly scan GPU, CPU, RAM, and memory bandwidth without requiring any installation or registration.The tool cross-references detected hardware specs against a catalog of approximately 60 open-source AI models from providers including Meta, Alibaba, Google, Microsoft, Mistral, DeepSeek, OpenAI, and NVIDIA, assigning each model a compatibility grade from S through F based on VRAM requirements, quantization options, and estimated inference speed. The entire detection process runs client-side, meaning no hardware data is sent to any server, and results appear within seconds of visiting the homepage.

• An AI agent platform called AllyHub launched to automate browsing, research, analysis, and reporting through plain-language task descriptions. The platform's AI agent, known as Ally, plans, executes, and delivers results via reusable workflows that handle web scraping, transcript extraction, sentiment and trend analysis, and scheduled monitoring. AllyHub designed Ally to learn while working, meaning repeated tasks become faster, more accurate, and cheaper over time. The platform targets marketers, analysts, creators, and teams seeking to eliminate repetitive knowledge-work and scale consistent, repeatable research and reporting without manual intervention.

• An AI-powered Excel agent called RowSpeak launched to convert messy spreadsheets, CSVs, PDFs, and image tables into cleaned data, charts, dashboards, and report-ready summaries using natural-language queries. RowSpeak built its AI to infer metrics, clean and normalize fields, detect trends and anomalies, and recommend and build visualizations automatically. The platform creates repeatable, reviewable workflows so nontechnical business users can ask questions, explain changes, and generate shareable reports without pivot tables or complex formulas. RowSpeak designed the tool to speed recurring reporting, cut manual data prep, and make spreadsheet analysis faster, more accurate, and auditable for finance, ops, marketing, and sales teams.

• Conversion of rough architectural previews into client-ready photoreal images and short videos became faster with the release of Rendervi, an AI render studio that processes model previews, sketches, and clay renders using image-to-image AI technology. Rendervi powered the platform with Nano-Banana Pro as its current rendering engine and integrated ChatGPT and Gemini as background models to analyze visuals and build prompts automatically. The studio delivers controlled inpainting, material swaps from a built-in texture library, environment tweaks, presets for cross-view consistency, and one-click upscaling. Rendervi created the tool for architecture teams who want to iterate faster while preserving design intent and avoiding manual rework.

  ---

• DEEP Robotics: DEEP Robotics Lynx M20S | The Ultimate All-Terrain Champion.

• ANYbotics: Vigier Ciment: Autonomous Robot Inspection in a 150-Year-Old Cement Plant

• AGIBOT: AGIBOT Livestream Day 1: Multiple AGIBOT G2 Humanoid Robots Working on a Real Tablet Production Line

• MindOn: One Model, Multiple Robots: Autonomous End-to-End Logistics in a Single Take

Welcome to this week's edition of Rushing Robotics! This week's lineup reads like a wishlist from a decade ago: electronic noses that can sniff out a single walnut fragment, quantum computers hitting fidelity numbers that make fault tolerance feel imminent, gene therapies rewiring brain development in mice, and robotic strawberry pickers that know exactly when fruit is ripe. We're watching multiple frontiers—biotech, quantum, photonics, robotics, and AI—advance simultaneously, and the cross-pollination between them is where things get genuinely interesting. Whether you're here for the deep tech, the product launches, or just to marvel at what's possible, you're in the right place.

Unlock truly uncensored AI—privately.

While most AI platforms store and analyze your conversations on their servers, Venice keeps everything local to your browser—meaning your private thoughts and queries stay truly private with no server-side storage or data harvesting.
Venice leverages Trusted Execution Environments (TEEs) for end-to-end encrypted processing—ensuring that even during computation, your data remains cryptographically protected from prying eyes.

Get a $10 credit when you register through my link. Upgrade to Venice Pro and apply the credit toward the most advanced models immediately. No filters, no logs—just pure creative freedom.

Claim your $10 credit & start creating!

[Disclosure: I earn a commission if you sign up via this link.]

🤯 Mind-Blowing

From Berkeley's carbon-nanotube e-nose detecting trace allergens, to China's first dedicated photonic computing lab, to Harvard's electrically-driven DNA synthesis chip cranking out 64 sequences in parallel—this section is all about breakthroughs that reframe what's technically feasible. Add in Quantinuum's 99.9975% one-qubit fidelity and WVU's soft robotic fruit harvester, and you've got a snapshot of how fast the "impossible" becomes "inevitable."

🔊 Industry Insights & Updates

Manufacturing and robotics take center stage this week. Oak Ridge's origami-inspired 3D printing slashes fabrication time by 95%. ABB and PSYONIC are teaching robots dexterity through prosthetic-hand data. Alibaba's Qwen-Robot suite bridges language models with physical action. And Divergent Technologies is scaling metal 3D printing for defense at unprecedented throughput. The common thread? Industrial-grade systems are getting faster, smarter, and more adaptable—fast.

🧬 BioTech

Three stories, three different angles on fighting disease. UNIFESP identifies SDC4 as a metastasis shield worth targeting. UC Riverside restores normal brain activity in Fragile X mice via gene therapy. And TU Munich's CellTrap chip lets researchers watch single immune cells hunt cancer in real time. Together, they hint at a future where treatments are both more precise and more personalized.

💡 Products/Tools of the Week

From Rendervi's AI render studio for architects, to Lium's conversational AI for messy real-world datasets, to NoteRich's privacy-first note-taking, to CC Switch's multi-provider AI coding CLI manager—this week's tools are about giving professionals more leverage without sacrificing control.

🎥 Video Section

Genesis AI introduces Eno, Faraday Future unveils its FF EAI robot ecosystem, and Boston Dynamics takes us inside the lab to see how Atlas learns. Three very different visions of embodied AI—worth watching back-to-back.

When photonic chips accelerate the AI that trains the robots that assemble the gene therapy delivery systems, we're not just progressing linearly; we're building compounding loops of innovation. The next decade will be defined less by what any one technology can do alone, and more by what happens when they start working together. Stay hungry, stay futurish!

🤯 Mind-Blowing

• Food safety monitoring could become more objective through an electronic nose created by researchers at the University of California, Berkeley, led by senior author Ali Javey and lead author Carla Bassil. The device uses 16 gas-sensitive carbon nanotube materials that react to different airborne molecules, addressing a longstanding manufacturing challenge where producing multiple sensing materials on a single chip has remained difficult. Carbon nanotubes form extremely thin conductive layers with large surface areas, making them highly sensitive to chemical compounds at room temperature and allowing the use of polymers that might degrade under heat. The system successfully identified as little as 0.05 grams of walnut material, roughly one-hundredth of an average shelled walnut, demonstrating potential for people with severe allergies. Bassil explained that the system uses the relative selectivity of gas sensors paired with machine learning pattern recognition to identify food-related chemical fingerprints more consistently than human smell alone, and she has already developed a portable version connecting to an iPhone application.

• A new research laboratory dedicated to photonic computing has been launched in Shanghai by China, marking the country’s first industry-academia platform focused specifically on light-based chip technology. The Shanghai Key Laboratory of Integrated Photonic Computing Chips and Systems, based at Shanghai Jiao Tong University, will bring together researchers and industry partners to develop photonic chips, optical components, computing architectures, and software systems. The facility is a joint initiative between Shanghai Jiao Tong University and Lightelligence, a Shanghai-based startup that has claimed to be the first company to achieve large-scale deployment of hybrid optical-electronic computing systems. Zou Weiwen, director of the new laboratory and a professor at Shanghai Jiao Tong University, described photonic computing as an important pathway for achieving breakthroughs in computing power, offering advantages in bandwidth, latency, and energy efficiency.

• A semiconductor chip capable of synthesizing 64 different DNA sequences in parallel using electric currents and a water-based enzymatic process has been developed by researchers at Harvard University, potentially offering an alternative to conventional DNA manufacturing methods. The chip uses localized electrical control to trigger DNA synthesis at selected sites on its surface, avoiding the solvent-heavy phosphoramidite chemistry widely used to produce synthetic DNA today. Each of the 64 synthesis sites contains two concentric ring electrodes surrounding DNA strands anchored at the center, with the inner electrode generating protons that lower pH to enable enzymatic DNA growth while the outer electrode consumes diffusing protons to prevent interference with neighboring sites. The approach generated 64 distinct DNA sequences simultaneously, with each sequence reaching up to 39 nucleotides in length, far exceeding the dozen sequences achievable with previous enzymatic methods.

• Fault-tolerant quantum computing has moved closer to reality after Quantinuum's Helios quantum computer achieved record fidelity levels of 99.9975 percent for one-qubit operations and 99.921 percent for two-qubit operations. The 98-qubit system with 50 logical qubits uses trapped-ion qubits combined with photonics, reducing risks to quantum computing data while significantly decreasing power consumption to as little as 40 kW. Quantinuum is working with Sandia National Laboratories under a Cooperative Research and Development Agreement to develop fault-tolerant computers, with SNL researchers developing a new benchmarking methodology that measures the performance of mid-circuit measurements, non-destructive readout operations that help correct quantum computing errors. Robin Blume-Kohout, a researcher at SNL involved in the work, said the laboratory evaluates every aspect of quantum computer performance with commercial partners to accelerate the advent of quantum supercomputing.

• A soft robotic gripper capable of detecting fruit ripeness and harvesting without damage has been developed by researchers at West Virginia University, led by assistant professor Anand Mishra from the Department of Mechanical, Materials and Aerospace Engineering. The device uses soft silicone and polyurethane fingers equipped with multiple sensors to measure a fruit’s size, shape, color and firmness before deciding whether it is ready to be picked. Tested on strawberries, the system removes fruit by twisting its stem rather than cutting it, addressing a major challenge in agriculture where delicate fruits have narrow harvest windows and bruise easily during picking, transportation and storage.

🔊 Industry Insights & Updates

• Fabrication time has been cut by 95 percent and costs reduced by 90 percent through a new origami-inspired 3D printing method developed by Steven Guzorek and fellow researchers at Oak Ridge National Laboratory. The mold-free process uses a high-strength fabric substrate such as nylon, glass fiber, or resin-infused fibers, combined with an integration layer of thermoplastic polyurethane to ensure adhesion. Reinforcing materials including thermoplastic carbon-fiber acrylonitrile butadiene styrene for lightweight strength or thermoset resins like epoxy for greater stiffness are then deposited, forming molecular-level bonds that create a robust, unified structure. Guzorek explained that the pioneering method redefines advanced manufacturing by fusing material science with transformative design principles, improving the efficiency and scalability of large-structure manufacturing while achieving forms unattainable with traditional additive approaches.

• A partnership has been formed between ABB Robotics and California-based bionic company PSYONIC to develop more dexterous robotic systems using real-world data from human prosthetic use. The collaboration combines PSYONIC's Ability Hand, a touch-sensitive prosthetic hand, with ABB's GoFa collaborative robot to study how human-generated touch and motion data can train robots to perform delicate and variable tasks. The initiative aims to address one of industrial robotics' biggest challenges—grasping and manipulation—while improving automation capabilities and potentially reducing engineering time for handling applications by up to 30 percent. Marc Segura, President of ABB Robotics, noted that human dexterity and the instinctive understanding of how to handle different objects remains one of the most difficult things to replicate in industrial-grade robotics.

• A first-of-its-kind embodied AI model family linking large language models with real-world robotic actions has been launched by Chinese firm Alibaba through its Tongyi Lab. The Qwen-Robot suite, currently undergoing pilot testing with selected Alibaba Cloud enterprise clients, comprises three specialized models targeting different layers of physical intelligence. Qwen-RobotNav focuses on movement and navigation, helping robots follow instructions, navigate to locations, track targets, and support autonomous driving. Qwen-RobotManip enables physical interaction such as grasping, moving, and manipulating objects using training data collected from different robotic systems. Qwen-RobotWorld acts as a world model that predicts how environments may change and helps robots understand likely outcomes of their actions, with the combined suite enabling machines to perceive, reason, and interact with the real world.

• Production throughput could roughly double compared with current-generation equipment through the Monolith One metal 3D printer developed by California-based Divergent Technologies. The machine offers a larger build chamber than many existing systems, allowing engineers to manufacture bigger and more complex structures in a single print. Divergent currently operates six Monolith One printers at its headquarters in Torrance, California, and plans to install an additional 64 systems at a new 430,000-square-foot facility in Long Beach over the next two years. Once fully operational, the site will support production of tens of thousands of munition airframes annually and have capacity to manufacture hundreds of thousands of critical metal components for aerospace and defense programs. The expansion comes as defense contractors seek faster production methods and more resilient domestic supply chains, with traditional manufacturing processes often requiring lengthy lead times for complex hardware.

Subscribe now

🧬 BioTech

• A molecule on the surface of cells that could serve as a promising therapeutic target against cancer has been identified by researchers at the Federal University of São Paulo (UNIFESP) in Brazil. The study, led by full professor Carla Cristina Lopes in the Department of Biological Sciences and supported by FAPESP, showed that overexpression of the Sindecam-4 (SDC4) receptor provides tumor cells with a protective shield against anoikis, a type of programmed cell death that occurs when cells detach from tissue. When researchers silenced the receptor using genetic engineering techniques, tumor cells lost their malignant properties and became dependent on adhesion to survive once again. The findings, published in the journal Cytotechnology, suggest that SDC4 could become both a therapeutic target for stopping metastasis before it begins and a diagnostic marker for monitoring disease progression.

• A gene therapy designed to replace a missing brain protein restored normal brain activity and improved behavior in a mouse model of Fragile X syndrome, according to a University of California, Riverside-led study published in Molecular Therapy Nucleic Acids. The research, led by senior author Iryna Ethell, a professor of biomedical sciences in the UCR School of Medicine, used a modified adeno-associated virus called AAV9 to deliver a healthy human version of the FMR1 gene into newborn mice lacking FMRP, the protein that regulates communication between brain cells. High-dose treatment produced significant improvements including normalized gamma brain-wave activity, reduced background neural noise, improved responses to sound, normal exploratory behavior, stronger social interactions, and improved cognitive flexibility. First author Anna Norman, a project scientist in Ethell's lab, and the research team found that timing was critical, with therapy administered during an early developmental period when the brain remains highly adaptable. The findings suggest gene therapy may one day address the underlying cause of FXS rather than simply treating its symptoms.

• Detailed observation of how individual immune cells interact with cancer cells has become possible through CellTrap, a microfluidic chip developed by Ghulam Destgeer and his research team at the Technical University of Munich's School of Computation, Information and Technology. The platform features 1024 small trapping chambers at the ends of continuously branching pathways from a main channel, into which cells are drawn and spatially fixed for observation over many hours. Unlike established laboratory tests that mainly capture average values across many cells, CellTrap reveals how each cell reacts and interacts with others, providing crucial timing information about when contact, activation, and killing of cancer cells occur. The researchers tested the platform on three cancer cell lines including glioblastoma, chronic myeloid leukemia, and adenocarcinoma, demonstrating that early activation signals in immune cells often predict later cell-damaging effects within the same interaction. Destgeer noted that the platform is not limited to immune and cancer cells, as almost any combination of cells can be loaded and observed in the chip.

  ---

💡Products/tools of the week

• Rendervi, an AI render studio, converts model previews, sketches, and clay renders into consistent, client-ready photorealistic images and short videos. The platform leverages image-to-image AI technology, currently utilizing Nano-Banana Pro, and integrates background models such as ChatGPT and Gemini to analyze visuals and build optimized prompts. Rendervi equips architecture teams with controlled inpainting tools, material swap options from a comprehensive texture library, environment tweak capabilities, and presets designed to maintain cross-view consistency across renderings. With one-click upscaling functionality included, the studio enables architects to iterate faster while preserving their design intent and significantly reducing manual rework throughout the creative process.

• Lium – AI for Complex Data, a conversational AI platform and infrastructure solution, connects, indexes, and profiles messy, terabyte-scale real-world datasets spanning geospatial, energy, space, infrastructure, and related fields. The platform automatically generates code, tools, and data transformations, provisions compute resources on demand, and allows teams to query and produce reusable analyses and outputs using natural language. Lium empowers domain experts to obtain fast, reliable insights without requiring heavy engineering involvement.

• NoteRich keeps users' original notes 100% local while using ephemeral, minimal-context cloud calls to deliver AI features like summarization, smart Q&A, scan & extract, memory-driven personalized insights, and local RAG indexing. The app provides users with powerful, evolving AI assistance along with cross-device P2P sync and encrypted storage, all without permanently uploading or exposing their full note archive. NoteRich makes modern AI productivity accessible to users who demand strict data control.

• CC Switch centralizes and visualizes configuration, provider keys, model routing, and failover for multiple AI coding CLIs including Claude, Gemini, Codex, Hermes, OpenCode, and OpenClaw, enabling developers to import presets, switch providers from the system tray, and avoid hand-editing JSON, TOML, or env files. The tool, developed by Jason Young (farion1231), runs a local API proxy, manages MCP servers and Skills, and tracks sessions and usage so comparing models, recovering from outages, and scaling multi-provider AI development becomes simpler and more reliable for developers and teams.

  ---

• Genesis AI: Meet Eno

• Faraday Future: Six-Series Full-Form FF EAI Robot World & FF EAI Education Ecosystem and Product Line Launch

• Boston Dynamics: How does Atlas learn? | Inside the Lab | Boston Dynamics

Welcome to this week's edition of Rushing Robotics, where the boundary between science fiction and reality dissolves a little more each week. As we cross into mid-June 2026, the pace of innovation feels less like progress and more like propulsion—each breakthrough building on the last in ways that redefine what we thought possible. From robots that read our muscles to chips that verify their own quantum integrity, this week's discoveries remind us that the future isn't arriving gradually. It's arriving in waves.

Unlock truly uncensored AI—privately.

While most AI platforms store and analyze your conversations on their servers, Venice keeps everything local to your browser—meaning your private thoughts and queries stay truly private with no server-side storage or data harvesting.
Venice leverages Trusted Execution Environments (TEEs) for end-to-end encrypted processing—ensuring that even during computation, your data remains cryptographically protected from prying eyes.

Get a $10 credit when you register through my link. Upgrade to Venice Pro and apply the credit toward the most advanced models immediately. No filters, no logs—just pure creative freedom.

Claim your $10 credit & start creating!

[Disclosure: I earn a commission if you sign up via this link.]

🤯 Mind-Blowing

This section features the kind of breakthroughs that make you reconsider what's possible. From wearable ultrasound systems that give robots human-like dexterity, to AI that learns physics before training, to quantum chips that verify their own integrity—these are the stories that push the boundaries of imagination. We also explore Hong Kong's first humanoid-run convenience store and a photomemristor that helps self-driving cars see through blinding light. The future isn't waiting for permission.

🔊 Industry Insights & Updates

The intersection of big tech and real-world deployment takes center stage here. NVIDIA and LG are weaving AI into factory floors and household robots, while French startup C12 advances quantum computing with carbon nanotube assembly. MIT's diamond-embedded chips could unlock 6G communications, and GigaAI has placed 100 humanoid robots into actual Chinese households for the first large-scale real-home test. The gap between lab and living room is closing fast.

🧬 BioTech

Life sciences are experiencing their own revolution. An AI model called TITO can now predict molecular evolution over 10,000 times faster than conventional methods, potentially transforming drug discovery. Zinc oxide nanowires are enabling liquid biopsies that capture cancer signals with 90% efficiency. And engineered extracellular vesicles are delivering full-length mRNA to restore dystrophin production in Duchenne muscular dystrophy—offering hope where viral vectors fell short. The code of life is becoming increasingly writable.

💡 Products/Tools of the Week

For builders and makers, this week brings some powerful new tools. Nous Research released Hermes Desktop, bringing the open-source Hermes Agent to macOS, Windows, and Linux with persistent memory, multimodal AI tools, and sandboxed execution. Archyl is redefining software architecture with AI-powered "architecture as code" that auto-generates interactive C4 diagrams and detects drift. TukiAI launched an AI sales and service agent for WooCommerce that syncs inventory in real-time and supports 54 languages across web, WhatsApp, and social channels. And Kuberns debuted an AI-powered cloud PaaS that automates end-to-end deployments, handling everything from CI/CD pipelines to cost optimization—no DevOps expertise required.

🎥 Video Section

See the future in motion: Google's Gemma Playground brings us a robot duck, UBTECH's U1 Pro humanoid robot blurs the line between human and machine, and Unitree's robots leave Simon Cowell speechless on America's Got Talent 2026. Sometimes you just have to see it to believe it.

There's something deeply thrilling about watching so many disparate threads—quantum hardware, embodied AI, biotech delivery systems, physics-aware neural networks—converge toward a future that feels both imminent and extraordinary. We're not just building smarter tools; we're building systems that understand the world's fundamental rules and operate within them more elegantly than we ever imagined. The question is no longer whether these breakthroughs will reshape our lives, but how quickly we'll adapt to a world where the impossible becomes routine. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A new wearable ultrasound wristband has been developed to give humanoid robots human-like dexterity by capturing detailed muscle and tendon activity beneath the skin. Researchers at MIT, led by Xuanhe Zhao, created the device to address the limitations of camera-based systems and strain sensors that struggle with continuous finger motions. The wristband uses a 256-channel wireless ultrasound imaging system and a hybrid Transformer-ResNet AI model to continuously track all 22 degrees of freedom of the human hand. In laboratory tests with eight volunteers, the system reproduced hand gestures with high accuracy and low latency, mirroring movements within 120 milliseconds and successfully recognizing all 26 letters of American Sign Language. The wireless technology allows users to remotely control robotic systems, which the team demonstrated by directing a robotic hand to play the piano.

• A machine learning system that learns the laws of physics before training has been developed to design advanced optical materials up to ten times faster than conventional methods. Researchers at Chalmers University of Technology embedded the fundamental laws of electromagnetism directly into the neural network, allowing the system to understand how light and electromagnetic fields behave from the start. This approach avoids forcing the model to discover these principles from scratch and significantly reduces the time required for data generation. According to Philippe Tassin, a professor at the Department of Physics and Astronomy, feeding the super-brain information about physics made it much smarter, reducing calculation times to one tenth of what was previously required.

• Hong Kong's first humanoid robot-run convenience store is set to open on the Hung Hom waterfront. A mainland Chinese firm specializing in embodied AI developed the 24-hour outlet, which features a robot nicknamed Xiao Gai that can interact directly with customers, initiate friendly conversations, and support purchases in multiple languages. The nine-square-meter capsule format can be configured to sell snacks, creative merchandise, and over-the-counter medicines. Financial Secretary Paul Chan revealed the project in his weekly blog, framing it as an endorsement of the city's open market and its role as a showcase for global innovation.

• A self-testing quantum random number generator chip has been developed to verify the integrity of its own hardware while producing random numbers. Researchers at the National University of Singapore, led by Associate Professor Charles Lim, created the device to eliminate the need to trust that measurement components are operating as specified. Using a measurement-device-independent protocol, the chip generates known quantum light states and compares detector responses against quantum theory predictions, stopping automatically if results do not match. The device integrates both the signal encoder and optical detector on a single silicon chip that operates at room temperature, achieving a detector efficiency of 69.1 percent and producing 64 certified random bits per second.

• A new photomemristor device mimicking the human eye has been developed to prevent self-driving cars from going blind in high-contrast lighting. Penn State researchers created the component to handle the transition from blinding brightness to deep shadow almost instantly, unlike the human eye which takes 20 to 30 minutes to acclimate. The device works by pairing titanium oxide with a stretchy plastic that swells in the dark and dries in bright light, acting as a physical volume knob for light sensitivity. This innovation allows autonomous vehicles to distinguish details like the glow of a red light against bright headlights, solving a major fragility in current robotic vision systems.

🔊 Industry Insights & Updates

• A connected AI ecosystem linking model development, simulation, and deployment has been unveiled through a broad collaboration between NVIDIA and South Korea's LG Group. The partnership aims to weave AI into factory floors, household robots, and autonomous vehicles by combining NVIDIA's accelerated computing platforms with LG's manufacturing and consumer electronics expertise. LG will use NVIDIA's digital twin technologies to create an autonomous manufacturing ecosystem that connects procurement, production, and logistics through real-time decision-making. For robotics, LG plans to incorporate NVIDIA Isaac Sim, Isaac Lab, and the GR00T foundation model to train home robots in virtual environments before deployment. The collaboration also extends to infrastructure, with LG Electronics working on liquid-cooling technologies for the NVIDIA DSX AI factory platform and LG Uplus building large-scale AI data centers.

• A nanoassembly process for transferring individual carbon nanotubes onto quantum chips has been advanced by French quantum computing startup C12 through a technique called pick-and-place. The company's Quantum Processing Unit integrates carbon nanotubes made from ultra-pure carbon-12 isotope on semiconductor chips, where these nanotubes house spin qubits. This development allows C12 to preselect and qualify individual carbon nanotubes before integration, enabling tighter quality control of assembled devices.

• A new chipmaking technique that embeds gallium nitride transistors into an ultrathin diamond layer has been developed by MIT researchers to improve high-power wireless systems. This method allows the diamond to act as a heat spreader within the chip, normalizing the temperature across the device so transistors can approach peak performance without reliability issues. The team built a power amplifier using this approach that achieved higher output power, efficiency, and gain than similar devices in existing literature. This advancement addresses the thermal management bottleneck that has previously limited the use of gallium nitride in demanding applications like 6G and satellite communications.

• A large-scale real-home test for general-purpose humanoid robots has been launched in China by Wuhan-based GigaAI, which deployed the first batch of 100 SeeLight S1 units into actual households. The robots successfully performed chores like preparing breakfast, loading dishwashers, and folding laundry after less than a month of on-site training. While the company utilizes an embodied foundation model that allows the machines to process natural language and adapt to changing environments, current limitations include slow execution speeds and difficulty handling delicate items like cups without spilling. GigaAI views this deployment primarily as a data-collection platform to improve future models.

Subscribe now

🧬 BioTech

• A new artificial intelligence model capable of predicting molecular evolution more than 10,000 times faster than conventional simulations has been developed to accelerate drug discovery. Researchers from Chalmers University of Technology and the University of Gothenburg created the TITO model to fast-forward through molecular dynamics, bypassing the need to calculate forces step by step over billions of femtoseconds. By learning the underlying rules of molecular motion from over 12,500 organic molecules and short peptides, the model can predict how new molecules will behave and change over time scales a thousand times longer than its training data. This breakthrough allows researchers to identify promising drug candidates more quickly and accurately without relying on computationally demanding numerical calculations.

• A new cancer detection device using zinc oxide nanowires has been developed to selectively capture extracellular vesicles from bodily fluids for liquid biopsies. The group, led by Nagoya University professor Takao Yasui and collaborating with Hokkaido University professor Yasuhide Inokuma, solved the challenge of attaching antibodies to nanowires by using a synthetic polymer variant called pKNHS 4.2. This specific polymer allows for single-step antibody modification, which enabled the device to achieve 90 percent capture efficiency for target molecules in cultured breast cancer cells. When tested on serum from ovarian cancer patients, the antibody-conjugated nanowires successfully isolated cancer-related vesicles and revealed distinct microRNA profiles, keeping the molecular contents intact for sensitive disease analysis.

• A new treatment platform using engineered extracellular vesicles has successfully delivered full-length messenger RNA to restore dystrophin production in preclinical models of Duchenne muscular dystrophy. Researchers at The University of Texas MD Anderson Cancer Center developed the natural nanoscale delivery particles to overcome the size limitations of current viral-based gene therapies, which cannot carry the entire DMD gene. The engineered vesicles were tagged to target skeletal muscles directly after bloodstream injection, resulting in improved muscle strength and function without the serious side effects or immune reactions typically associated with viral vectors. This breakthrough suggests that extracellular vesicle-mediated mRNA therapeutics could serve as a blueprint for restoring large proteins in other degenerative diseases.

  ---

💡Products/tools of the week

• Nous Research released Hermes Desktop, bringing the open-source Hermes Agent to macOS, Windows, and Linux with a polished chat interface. The application shares persistent memory, sessions, API keys, and skills with the CLI and web gateway, ensuring work context syncs across all interfaces. Users can invoke multimodal AI tools like web search and image generation, schedule recurring automations, and spawn isolated subagents for delegated tasks. It supports various backends including local, Docker, and Modal for hardened, sandboxed execution.

• Architecture as code is now possible through a new AI-powered platform. Archyl released a living software architecture documentation tool that uses AI to analyze codebases and automatically discover components. It generates interactive C4 diagrams and detects drift so documentation remains accurate. By expressing architecture as code via a YAML DSL, Archyl enables integration with Git workflows for versioning, reviews, and collaboration. Architects, tech leads, and dev teams can now replace manual, stale diagrams with faster, more accurate, and reviewable architecture docs.

• Real-time inventory synchronization and multilingual customer support are now available for WooCommerce stores. TukiAI released an AI sales and service agent as a WordPress plugin that leverages leading LLMs such as GPT-4, Claude, Gemini, and Mistral to read catalogs, sync prices, and manage orders. The platform operates 24/7 across websites, WhatsApp, and social channels, answering questions in 54 languages and offering context-aware product recommendations. TukiAI programmed the agent to escalate to humans when unsure, helping store owners automate repetitive support, recover abandoned carts, boost conversions, and free their teams for higher-value tasks.

• Kuberns launched an AI-powered cloud PaaS that uses agentic AI to automate end-to-end deployments and operations. The platform auto-detects applications, generates configs, and wires CI/CD pipelines automatically. Kuberns handles scaling, monitoring, secrets, and cost optimizations so teams can ship in minutes without needing DevOps expertise. Users leverage the platform to cut deployment time, reduce cloud spend, and access production-grade automation like live logs and alerts with minimal setup.

  ---

• Google for Developers: Gemma Playground: Robot Duck

• DPCcars: UBTECH U1 Pro Humanoid Robot Is Shockingly Real

• America’s Got Talent: Unitree’s Robots Leave Simon Cowell SPEECHLESS! | Auditions | AGT 2026

Welcome to this week's edition of Rushing Robotics! We are living through one of the most remarkable periods in human history — a time when the boundaries between science fiction and scientific fact dissolve a little more with every passing week. This edition is packed with stories that would have seemed impossible just a decade ago: a sticker that steadies a broken heart, robots teaching children to walk again, and a leaf made of silicon that breathes in carbon dioxide and breathes out fuel.

Unlock truly uncensored AI—privately.

While most AI platforms store and analyze your conversations on their servers, Venice keeps everything local to your browser—meaning your private thoughts and queries stay truly private with no server-side storage or data harvesting.
Venice leverages Trusted Execution Environments (TEEs) for end-to-end encrypted processing—ensuring that even during computation, your data remains cryptographically protected from prying eyes.

Get a $10 credit when you register through my link. Upgrade to Venice Pro and apply the credit toward the most advanced models immediately. No filters, no logs—just pure creative freedom.

Claim your $10 credit & start creating!

[Disclosure: I earn a commission if you sign up via this link.]

🤯 Mind-Blowing

Some breakthroughs don't just improve existing technology—they fundamentally reimagine it. This section highlights innovations that challenge our deepest assumptions about what's possible, from medical devices that replace invasive surgery with wearable stickers to quantum light that multiplies laser power twentyfold without the damage. These are the kinds of advances that make you pause and reconsider the boundaries of human capability.

🔊 Industry Insights & Updates

The gap between laboratory promise and real-world deployment is shrinking by the day. This week, the biggest names in technology and logistics are making bold moves—Amazon's robots are learning to take conversational orders, NVIDIA is betting big on physical AI with a new omnimodel, and China's postal system is putting humanoid workers on one of the busiest sorting floors in the world. Meanwhile, new research reveals that consistency might be the secret ingredient teaching robots real dexterity. The robotics revolution isn't coming—it's already here, reshaping industries in real time.

🧬 BioTech

In the fight against humanity's most devastating diseases, hope often comes in measured doses. This week brings more than measured progress: a drug that doubles survival for metastatic pancreatic cancer patients, a genomic test that could spare thousands from the grueling side effects of unnecessary chemotherapy, and a new immunotherapy combination that may finally overcome treatment resistance in solid tumors. These aren't incremental advances—they're potential turning points that could redefine standard care and give patients something priceless: more time with the people they love.

💡 Products/Tools of the Week

Four tools this week that put serious automation power into the hands of non-technical users. Minded converts browser workflow recordings into AI agents that understand context rather than just repeating clicks — train them with plain language or a visual editor, no API required. Anchor Browser brings a fully humanized, cloud-hosted Chromium environment to AI agents, handling authentication, proxies, and CAPTCHAs while integrating with frameworks like LangChain. Twin goes a step further with a no-code platform that transforms plain-English instructions into always-on web automations that self-heal when websites change. And Floor Plan AI rounds things out by generating presentation-ready 2D floor plans and downloadable 3D models from a simple text prompt or reference image — no CAD skills needed.

🎥 Video Section

Sometimes the future needs to be seen to be believed. This week's video selections bring you face-to-face with the humanoid robots stepping off the assembly line and into our world—from upgraded performance models to the global summits where industry leaders debate our shared future with machines. Watch as the line between prototype and coworker continues to blur.

The convergence of AI, robotics, and biotechnology is accelerating at a breathtaking pace, and each week brings us closer to a world that would have seemed like science fiction just a decade ago. What excites me most isn't just the individual breakthroughs—it's the way they compound, building on each other to create possibilities we can barely imagine today. A wearable pacemaker here, a quantum laser technique there, a cancer drug that doubles survival—these aren't isolated milestones. They're threads in a tapestry that's weaving a future of unprecedented human potential. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A noninvasive, surgery-free alternative to standard cardiac implants has been developed by MIT researchers in the form of a wearable ultrasound pacemaker sticker. The technology uses focused ultrasound waves to control the rhythm of a beating heart by pairing an external acoustic sticker with sonogenetics, a field where genetically engineered cells respond to sound. A one-time gene therapy injection modifies heart cells with custom ion channels that open when hit by specific acoustic frequencies, allowing a rush of calcium to activate the cell. When tested on engineered human cardiac cells and live rats with arrhythmias, the sticker successfully steadied irregular heartbeats and brought slow hearts back to a healthy pace.

• A lightweight robotic device has enabled children with spinal muscular atrophy to stand up from a chair unassisted for the first time. Researchers from Beihang University, Peking University Third Hospital, and the Massachusetts Institute of Technology developed the knee-mounted robot which weighs just 2.1 pounds. Instead of assisting movement, the device applies controlled resistance during video game-based training to strengthen weakened muscles. After six weeks of training, six children with SMA Type II showed a 130 percent increase in muscle strength and a 19 percent increase in quadriceps size, allowing all participants to rise from a seated position independently.

• The world’s first tower crane 3D construction printer has been unveiled by Luyten, an Australian robotics company. Named the Ascend Series, the machine integrates traditional tower crane architecture with robotic concrete printing to build structures up to 328 feet tall. The system tackles labor shortages and material waste by using artificial intelligence to generate print paths and monitor progress in real time. According to Luyten founder and CEO Ahmed Mahil, the technology transforms one of the construction industry’s most important machines into a robotic manufacturing system that builds directly from digital designs.

• A solar-powered device that transforms carbon dioxide and water into methanol without external electricity has been developed by researchers at Yale University. The artificial leaf system relies entirely on sunlight to drive the chemical reaction, marking a major step forward in artificial photosynthesis technology. By combining a specialized catalyst that handles a complex six-electron reaction with a redesigned photoelectrode made of microscopic silicon pillars, the team created one of the most efficient silicon-based methanol conversion devices reported so far. Yale chemistry professor Hailiang Wang noted that the concept is comparable to what nature does, and the breakthrough could strengthen future efforts to capture atmospheric carbon while producing cleaner liquid fuels for transportation and industry.

• A 20-fold enhancement in ultrafast laser interactions has been demonstrated without increasing overall laser power. Researchers led by Jian Wu at East China Normal University achieved this boost by utilizing a form of quantum light known as bright squeezed vacuum, which creates extreme fluctuations in photon density to produce short-lived bursts of high instantaneous intensity. The team found that a bright squeezed vacuum pulse containing just 300 nanojoules of average energy produced the same nonlinear ionization effect in sodium atoms as a conventional laser pulse with more than 20 times the effective intensity. This approach, detailed in the journal Nature, reduces the risk of thermal or structural damage to materials, tackling a major limitation in modern laser physics where nonlinear effects typically require intense pulses that harm the systems being studied.

🔊 Industry Insights & Updates

• A new version of the autonomous Proteus robot capable of taking plain-language instructions from workers has been unveiled by Amazon. Introduced at the Delivering the Future event in London, the upgraded robot can now move items throughout fulfillment sites instead of being limited to dock operations. Employees can direct the machine using conversational text prompts, allowing the robot to determine task priority, select routes, and decide when to act. Amazon is currently piloting the new Proteus in its research facilities and plans to begin deployment across Europe in the first half of 2027 as part of a broader effort to combine robotics with artificial intelligence inside its warehouses.

• New physical AI technologies were announced by NVIDIA at GTC Taipei, headlined by Cosmos 3, an open foundation model that the company describes as the first fully open omnimodel capable of understanding and generating text, images, video, ambient sound, and actions within a single system. The model is built on a mixture-of-transformers architecture combining reasoning and content generation, and NVIDIA claims it tops several open-model benchmarks for world generation, robotic action policies, and vision understanding. Jensen Huang, founder and CEO of NVIDIA, stated that the big bang of physical AI is just around the corner thanks to breakthroughs in multimodal reasoning language, vision and world models, and that Cosmos 3 gives developers a generational leap in ability to build robots, autonomous vehicles and vision AI that perceive, reason, plan and act in the physical world.

• Consistent training data has been proven to teach robots dexterity much better than highly variable examples. Researchers from New York University Tandon School of Engineering and the Robotics and AI Institute discovered that robots trained with structured, predictable demonstrations performed significantly better on complex manipulation tasks. The team found that popular motion-planning algorithms called rapidly exploring random trees produced too much variation, confusing the imitation learning system. By developing alternative planning approaches that prioritized steady progress or used predefined motions, the researchers generated more consistent data. When tested on tasks like rotating a cylinder with two arms or manipulating a cube with a robotic hand, the robots trained on consistent demonstrations achieved near-perfect performance in simulation and successfully transferred those skills to physical hardware with high success rates.

• A deployment of humanoid robots into one of the world’s busiest postal networks has been completed by China as the country accelerates efforts to automate logistics at a massive scale. Images released from the Jianggao logistics site under the Guangzhou postal center show the humanoid robotic sorters handling parcels alongside robotic arms and unmanned forklifts. According to state media, the robots are capable of processing up to 1,200 parcels per hour as part of a broader modernization push by China Post Group. The facility can allegedly handle an average of 6.5 million pieces of mail every day, with peak volumes exceeding 10 million, prompting operators to increasingly turn toward autonomous systems capable of working continuously with minimal human intervention.

Subscribe now

🧬 BioTech

• A near doubling of overall survival for metastatic pancreatic cancer patients was achieved by the experimental drug daraxonrasib in a pivotal Phase 3 clinical trial conducted at the Dana-Farber Cancer Institute. Patients receiving daraxonrasib lived a median of 13.2 months compared to 6.7 months for those on second-line chemotherapy, representing a 60 percent reduction in the risk of death. The RASolute 302 trial also showed the drug delayed disease progression, with median progression-free survival reaching 7.2 months versus 3.6 months with chemotherapy. As an oral RAS(ON) multi-selective inhibitor designed to block signaling from both mutant and non-mutant RAS proteins, daraxonrasib produced higher tumor response rates regardless of mutation status, findings that were published in The New England Journal of Medicine.

• A genomic test has been shown to help thousands of breast cancer patients safely avoid chemotherapy without compromising their outcomes, according to results from a large international phase III clinical trial. Researchers led by University College London and supported by the UK’s National Institute for Health and Care Research conducted the OPTIMA trial, which enrolled more than 4,400 patients with hormone receptor-positive, HER2-negative early breast cancer. The study utilized the Prosigna Breast Risk of Recurrence test, developed by Veracyte, which analyzes the activity of 50 genes in a tumor sample to estimate the risk of cancer returning. Findings presented at the American Society of Clinical Oncology Annual Meeting revealed that outcomes among lower-risk patients were nearly identical regardless of whether they received chemotherapy, demonstrating that many can skip the treatment and its severe side effects while relying on hormone therapy alone.

• Promising results for overcoming immunotherapy resistance have been highlighted in a new comprehensive review published in the Journal of Pancreatology. The review identifies TIGIT as a key second-generation immune checkpoint target that addresses the limitations of first-generation inhibitors like drug resistance and low response rates. Unlike other emerging checkpoints, TIGIT is widely expressed in T cells, natural killer cells, and regulatory T cells, allowing it to suppress immune funPromising results for overcoming immunotherapy resistance have been highlighted in a new comprehensive review published in the Journal of Pancreatology.ction through multiple pathways. Researchers found that combining TIGIT inhibitors with PD-1 inhibitors, such as tiragolumab with atezolizumab, significantly improves objective remission rates and progression-free survival compared to monotherapy. This combination works by reversing T-cell exhaustion and preventing natural killer cell depletion, offering a new translational framework to treat solid tumors and hematologic malignancies.

  ---

💡Products/tools of the week

• Minded has introduced a platform that uses artificial intelligence to convert browser workflow recordings into adaptable automation agents. By pairing large language models with a screen recorder, the system captures user interactions and generalizes them into agents that understand context rather than merely repeating clicks. Users can train these agents through natural language or a visual editor called "vibe-coding," and developers can export them for deeper customization. The agents interact directly with user interfaces, eliminating the need for API integrations while still being able to handle process variations, comprehend documents, and execute multi-step instructions.

• A new cloud-hosted Chromium platform launched that allows AI agents to reliably drive real web browsers at scale. Anchor Browser introduced this humanized browsing environment to automate complex UI flows by using AI to plan and compile deterministic browser workloads, only invoking runtime LLM steps when absolutely necessary. The platform handles authentication, proxies, and CAPTCHAs seamlessly while integrating with popular agent frameworks like LangChain. Teams dealing with sites that lack APIs now have a secure, scalable, and low-fragility solution for their browser automation needs.

• A new no-code AI agent platform emerged that transforms plain-English instructions into autonomous, always-on automations. Twin created this solution to connect to APIs and control browsers, enabling tasks like logging in, clicking, scraping, and filling forms without manual intervention. The system uses artificial intelligence to design the steps, handle exceptions, and self-heal when websites change, ensuring workflows remain robust. It also selects the best models for each task, allowing non-technical teams and enterprises to automate repetitive web- and app-based work at scale and cut costs while getting production-ready results without writing code.

• A new AI-powered generator debuted that converts plain-English prompts or reference images into presentation-ready 2D floor plans and downloadable GLB 3D models. Floor Plan AI created this tool to automatically lay out rooms, walls, doors, and dimensions while applying selectable drawing and interior styles. The system produces PNG and GLB exports in seconds to minutes, providing homeowners, real-estate teams, and designers with fast, no-CAD concept layouts and ready-to-use visual assets for listings, renovations, or 3D pipelines.

  ---

• DEEP Robotics: DR02 Humanoid Robot | Performance Upgraded. Possibilities Expanded.

• Financial Times: Are consumers ready for humanoid robots?

• ANNnewsCH: Asia's First "Humanoid Summit" Kicks Off, Featuring Robots to Help Solve Labor Shortages

Welcome to this week's edition of Rushing Robotics! The future doesn't wait — and neither does this week's edition. From hearts grown in a lab to robots rolling off assembly lines every 15 minutes, the pace of innovation is accelerating faster than ever. Whether you're a tech enthusiast, a curious mind, or someone who simply wants to stay ahead of the curve, you're in the right place. Let's dive into the breakthroughs, updates, and tools shaping the world of tomorrow.

Unlock truly uncensored AI—privately.

While most AI platforms store and analyze your conversations on their servers, Venice keeps everything local to your browser—meaning your private thoughts and queries stay truly private with no server-side storage or data harvesting.
Venice leverages Trusted Execution Environments (TEEs) for end-to-end encrypted processing—ensuring that even during computation, your data remains cryptographically protected from prying eyes.

Get a $10 credit when you register through my link. Upgrade to Venice Pro and apply the credit toward the most advanced models immediately. No filters, no logs—just pure creative freedom.

Claim your $10 credit & start creating!

[Disclosure: I earn a commission if you sign up via this link.]

🤯 Mind-Blowing

Scientists have grown a functioning biological pacemaker from stem cells, NVIDIA is closing the gap between virtual training and real-world robotics, a $5 chip is detecting early-stage lung cancer with 94.9% accuracy, a living bandage is revolutionizing wound healing, and humanoid robots are now rolling out of a Shenzhen factory every 15 minutes.

🔊 Industry Insights & Updates

The industrial world is getting a serious AI upgrade. Google and FANUC are teaming up to bring Physical AI into factories, perovskite solar cells are pushing past 33% efficiency, Bosch is backing humanoid robots for logistics, and Anthropic just launched Claude Opus 4.8 — a model built to be radically more honest when it matters most.

🧬 BioTech

Biology is becoming programmable. This week, researchers unveiled an intelligent tattoo that detects skin cancer four days after tumors form, a seed-sized surgical robot that performs five medical tasks in under a second, and a new microscopy technique that reconstructs 3D images of brain tissue using half the light — and half the time.

💡 Products/Tools of the Week

Your workflow just got smarter. This week's tool picks cover the full stack of modern productivity — from an AI software engineer that autonomously opens pull requests, to a Chrome extension that turns your saved links into structured knowledge, an AI filmmaking studio for production-grade video, and an open-source infinite canvas that thinks alongside you. These aren't just tools; they're upgrades to how you work, create, and learn.

🎥 Video Section

See it to believe it. This week's video picks put the hardware front and center — from LimX's next-gen humanoid Luna to Boston Dynamics showing off football fundamentals, Engineai's massive Shenzhen manufacturing base going live, and Astribot's T1 making its debut.

We're living through a moment where biology, robotics, and artificial intelligence are converging at a speed that would have seemed like science fiction just a decade ago. Biological pacemakers, cancer-sniffing chips, and robots that learn in simulation before stepping into the real world aren't distant promises — they're happening right now, this week. The tools being built today will redefine medicine, manufacturing, and human potential in ways we're only beginning to imagine. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A lab-grown biological pacemaker capable of mimicking the heart’s natural rhythm control system has been developed by scientists in Shanghai. By using human pluripotent stem cells, the team from the Chinese Academy of Sciences and Fudan University created a three-dimensional sinoatrial node organoid that generates electrical impulses. The researchers linked this organoid to an artificial cardiac plexus to recreate how the nervous system communicates with the heart. This lab-grown tissue produced stable and spontaneous beating while closely matching human embryonic sinoatrial node cells in gene activity. It also reacted correctly to medications used to control heart rate. The findings were published in the journal Cell Stem Cell and could pave the way for biological pacemakers that reduce reliance on electronic devices.

• Robots trained entirely in simulation are performing more reliably in the real world, according to new research presented by NVIDIA at the International Conference on Robotics and Automation (ICRA). The studies focus on narrowing the sim-to-real gap where machines trained in virtual environments struggle outside controlled lab conditions. The COMPASS framework trains robots in NVIDIA Isaac Lab simulations before transferring skills to different physical bodies, achieving about 80 percent success across 20 real-world navigation trials. For grasping in cluttered environments, the Grasp-MPC system adjusts its motion in real time and achieved a 75 percent success rate on unfamiliar objects after training on two million simulated trajectories.

• A new handheld cancer-screening device capable of detecting early-stage cancer biomarkers from a single drop of blood has been developed by researchers at Westlake University. The technology shrinks refrigerator-sized laboratory equipment into a portable system that boosts detection accuracy to about 10,000 times that of conventional methods. It utilizes a 3D Bound States-in-the-Continuum sensing chip that requires only an LED light source and a photodetector to operate. By shifting the focus from complex light wavelengths to simple light intensity through Q-modulated refractometric sensing, the device can detect microscopic changes in how light bends when cancer biomarkers are present. The team also utilized an innovative aluminum-based fabrication technique akin to movable-type printing, allowing for the mass production of thousands of identical chips on a single wafer and slashing the cost per chip to just $5. In trials with 171 patient serum samples conducted in partnership with Xiamen University, the device achieved 94.9% accuracy for early lung cancer detection and 92.1% for post-surgery monitoring, compared to just 74.7% for the standard laboratory method.

• A breakthrough in treating chronic wounds has been achieved with the development of a cytokine factory patch that acts as a living bandage, created by researchers at Rice University. The patch functions as a cell-based delivery platform using engineered ARPE-19 cells encapsulated inside a protective biocompatible matrix to continuously produce signaling proteins such as IL-10, IL-12, and TGF-β. Because healing proteins in existing therapies degrade quickly and fail to remain at the wound site, the Rice team designed the matrix to shield the engineered cells from the host immune system while allowing nutrients and therapeutic proteins to move through the material. This setup sustains localized cytokine delivery for longer periods, maintaining a stable healing environment during recovery. In preclinical studies, the patch accelerated healing in excisional wound models in mice and pigs. RNA sequencing revealed activation of several pathways associated with wound healing, showing increased activity in genes linked to tissue regeneration and immune modulation. The platform offers flexibility beyond a single treatment strategy, as scientists can modify the engineered cells to produce different combinations of cytokines, growth factors, or therapeutic proteins for tailored treatments.

• High-volume production of humanoid robots is now underway at a new 129,000-square-foot smart manufacturing facility opened by ENGINEAI in Shenzhen. The plant utilizes a fully integrated manufacturing workflow that enables the production of one humanoid robot every 15 minutes. The first batch of the company's T800 heavy-duty humanoid robots has already been completed at the Honghualing district location. To maintain quality during large-scale production, every robot undergoes 79 quality checks and 46 simulation tests prior to delivery. ENGINEAI stated the new facility is designed to boost efficiency and support quicker commercialization of its robots for industrial and inspection applications.

🔊 Industry Insights & Updates

• A new collaboration has been formed to advance Physical AI in industrial robots. Google and FANUC America Corporation are joining forces to combine artificial intelligence with industrial robotics systems already used worldwide. The partnership aims to deploy AI-powered robots capable of understanding their surroundings through sensors, making autonomous decisions, and carrying out complex tasks in real-world factory environments. FANUC has already shipped more than 1,000 robots for Physical AI-related applications since showcasing the platform last December. The collaboration reflects a broader industry shift toward adaptive factory automation systems capable of handling labor shortages and more flexible manufacturing operations.

• A targeted passivation technique has been developed to significantly improve the performance of perovskite/silicon tandem solar cells. Researchers from the Ningbo Institute of Materials Technology and Engineering, Soochow University, and Taizhou University devised a peak-selective passivation strategy to mitigate electrical leakage caused by uneven perovskite layer deposition on industrial silicon substrates. By using polystyrene nanospheres as a template to apply a thin insulating layer of aluminum oxide specifically to the peaks of the silicon pyramids, the team blocked leakage pathways without interfering with charge transport. Testing on a one-square-centimeter device yielded a power conversion efficiency of approximately 33 percent, with the cell retaining roughly 90 percent of its initial efficiency after 1,000 hours of continuous operation.

• A new partnership has been formed to scale the production of industrial humanoid robots. London-based robotics company Humanoid and German firm Bosch are joining forces to accelerate the deployment of the HMND platform across logistics and manufacturing sectors. The collaboration follows a successful proof of concept in March 2026 where robots moved boxes of five different sizes from a conveyor belt to a trolley in Germany. Bosch will provide strategic guidance and technical expertise in hardware design, production, and supply chain management to help transition the robots from prototype validation to industrial-scale deployment. The HMND 01 platform comes in two configurations, a bipedal model standing 5 feet 10 inches tall and a larger wheeled variant measuring 7 feet 3 inches, both capable of carrying payloads of up to 33 pounds in human-centric industrial environments.

• A new flagship AI model designed to be more honest during complex tasks has been launched by Anthropic. Claude Opus 4.8 is four times less likely than its predecessor to overlook coding flaws and is better at flagging uncertainty instead of silently passing errors. The upgrade also introduces Dynamic Workflows for Claude Code, allowing hundreds of parallel AI subagents to handle large codebase migrations. Anthropic has additionally added an effort control setting for users to manage computational reasoning time and reduced pricing for its fast mode. The model is available now through the Claude API and claude.ai.

Subscribe now

🧬 BioTech

• A new experimental platform called SMEAR-ULM has been developed by Canadian researchers to detect melanoma before the cancer becomes visible on the skin. The system identifies tiny heat changes produced by early-stage tumors using a combination of microneedles, nanoparticles, and ultrafast imaging. A small patch of painless microneedles deposits rare-earth nanoparticles just beneath the skin to create a temporary intelligent tattoo that acts as microscopic temperature sensors. When near-infrared light shines on the area, the nanoparticles emit visible light with a duration that changes based on local temperature, allowing the system to capture the response in a single snapshot and convert it into a detailed thermal map. The researchers from the Institut national de la recherche scientifique and Université de Montréal say the system detected micro-melanomas just four days after they formed in laboratory models, far earlier than conventional thermal imaging systems that typically only catch tumors larger than five millimeters.

• A new seed-sized surgical robot that can perform five distinct medical tasks in less than a second has been created to advance minimally invasive procedures by researchers at Nanyang Technological University (NTU) Singapore. Measuring just 4.4 millimeters long, the device is controlled wirelessly through weak magnetic fields and can move across soft surfaces, cut biological tissue, release drugs, grip and store tissue samples, and generate heat remotely. Unlike most miniature magnetic robots that perform only one or two tasks, this invention uses a magnetic module that can be magnetized, demagnetized, and remagnetized in different directions to activate different functions. The team, led by Associate Professor Lum Guo Zhan, built the robot from soft silicone-based materials known as PDMS and Ecoflex, with microscopic magnetic particles embedded throughout. The researchers also engineered different regions to respond independently to magnetic signals and added a sixth degree of freedom known as rolling, allowing the robot to spin around its long axis for better navigation in narrow spaces. The findings were published in the journal Advanced Materials.

• A new microscopy technique called AIMED has been developed by researchers that reconstructs three-dimensional images from significantly fewer measurements, potentially accelerating biological imaging while reducing light exposure to samples. The approach, which stands for Arbitrary Illumination Microscopy with Encoded Depth, combines optical encoding with computational image reconstruction for multiphoton microscopy applications. Rather than scanning each depth individually as conventional methods do, the system excites multiple layers simultaneously and computationally separates the signals using sparse optimization algorithms. The technique uses a spatial light modulator to split a laser beam into multiple focal points positioned at different depths, with independently adjustable intensity to compensate for signal loss deeper inside tissue. Testing on mouse brain neuronal samples demonstrated that AIMED could resolve fine structures such as dendrites and axons while operating at a compression ratio of roughly 60%, requiring only one-half to one-third of the optical power normally used per imaging plane.

  ---

💡Products/tools of the week

• Automating the entire workflow from reading tickets to opening pull requests is the core function of Codowave, an autonomous AI software engineer. The platform connects to existing repositories and issue trackers, maps the full repository context, and writes a scoped plan before generating and self-reviewing code. It runs tests and opens pull requests with attached evidence and rationale. Codowave uses large language models, including Claude and Codex, to reason about changes and adhere to team conventions. Teams can safely offload routine fixes and small features to increase development velocity while ensuring work remains inside existing PR and CI review workflows.

• Building a continuously evolving second brain that reduces information overload is the core function of Knowly, a Chrome extension and personal knowledge platform. Powered by a proactive AI agent, Knowly captures anything a user saves, including webpages, PDFs, videos, and tweets, and automatically clusters those sources into topics. The system synthesizes the information into GenPages and multi-page Flows, creating guided learning journeys tailored to depth without relying on prompts or chat logs. By continuously reading the saved corpus and browsing behavior, Knowly surfaces digests, cross-source connections, suggested next steps, and on-the-fly topic pages. Users can automate organization, get personalized explanations, and move from saved links to structured understanding with minimal friction to speed learning.

• Scaling visual output and simplifying the end-to-end production lifecycle is now achievable for creators, boutique studios, and production teams using Birdwalk. This AI filmmaking platform lets users generate, edit, manage, and collaborate on production-grade video and image assets by orchestrating top generative models such as Seedance 2.0, Veo 3.1, and Kling through a VFX-designed pipeline. The system includes built-in editing for color, crop, and rotate, plus automatic versioning, shot generation, and sequence assembly. Embedding AI directly into generation and editing workflows helps speed iteration, standardize quality, and keep teams perfectly in sync throughout the project.

• Unifying notes, sketches, code, and live widgets on a single board is the core function of Dim0, an open-source, privacy-first infinite thinking canvas. A board-aware AI agent reads the workspace to search and synthesize information, runs code, and drops editable results like mind maps, summaries, charts, and nodes directly onto the canvas. Featuring multi-model support and optional self-hosting, Dim0 gives researchers, designers, engineers, and creators contextual AI that acts on their actual work. Users can consolidate tools, keep spatial structure in their thinking, and retain complete control over their data.

  ---

• LimX Dynamics: Meet LimX Luna—Our Next-Gen Full-Size Interactive Humanoid Robot.

• Engineai Robot: The Engine AI Intelligent Manufacturing Base in SHENZHEN is officially commissioned.

• Boston Dynamics: School of Football | The Basics | Boston Dynamics x Hyundai

• Astribot: Astribot T1 is here. Create freely. Shine anywhere.

Welcome to this week's edition of Rushing Robotics! This week's newsletter is packed with innovations that sound like they're ripped from a sci-fi novel—but they're very real. From stretchable hydrogel implants that talk to our arteries, to humanoid robots auctioning off to the highest bidder, to microrobot swarms you can control with light, the future is arriving faster than ever. Let's break it down.

Unlock truly uncensored AI—privately.

While most AI platforms store and analyze your conversations on their servers, Venice keeps everything local to your browser—meaning your private thoughts and queries stay truly private with no server-side storage or data harvesting.
Venice leverages Trusted Execution Environments (TEEs) for end-to-end encrypted processing—ensuring that even during computation, your data remains cryptographically protected from prying eyes.

Get a $10 credit when you register through my link. Upgrade to Venice Pro and apply the credit toward the most advanced models immediately. No filters, no logs—just pure creative freedom.

Claim your $10 credit & start creating!

[Disclosure: I earn a commission if you sign up via this link.]

🤯 Mind-Blowing

Penn State's CaroFlex hydrogel implant eliminates stitches and lowers blood pressure by 15%. UPenn created hybrid light-matter particles that could revolutionize optical computing. JD.com is auctioning humanoid robots. Horizon Robotics released HoloMotion-1, an open-source model controlling humanoid movements at 300 fps. And Boston Dynamics' Atlas now lifts 100+ pound refrigerators thanks to millions of hours of simulation training.

🔊 Industry Insights & Updates

A neuromorphic event camera from the University of Arizona solves 3D tracking in mixed-reflectivity scenes. Menlo Research unveiled the Asimov DIY humanoid kit for independent developers at $15,000. Hyundai is deploying 25,000 Atlas robots across US manufacturing plants by 2028. And Yokohama National University developed recyclable 3D printing resin that can be melted and reused 10 times.

🧬 BioTech

Light-controlled microrobot swarms made from microalga assemble into custom shapes and deliver drugs with 90% transfer rates. A new organ-on-a-chip model recreates IBD and reveals that stromal fibroblasts actively drive the disease. And Argonne National Laboratory launched RoSA, an initiative to create AI robotic lab assistants that learn by watching human scientists.

💡 Products/Tools of the Week

Plurai introduced vibe training to generate data and deploy reliable AI agent models in minutes without labeled data. Tycoon.us launched an AI CEO named Astra that manages over ten ready-to-use AI agents to run your company based on simple KPIs. WeWeb helps builders finish AI-generated apps, letting you take full control with a no-code editor to push past the typical 80% completion limit. And OpenHuman provides a local-first, privacy-first AI agent that remembers everything over time, all in one simple open-source interface.

🎥 Video Section

Genesis AI's GENE operating autonomously, Xynova's Flex 2 dexterous hand, ANYbotics' industrial inspection robots, and ORBIT Robotics' HELIOS in action.

The pace of innovation right now is staggering. Every week brings breakthroughs that fundamentally reshape what's possible—from our arteries to our factories to our laboratories. We're watching the future being built in real-time, and the most exciting part is that this is just the beginning. The tools, robots, and medical marvels emerging today will define how we live, work, and heal tomorrow. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A new bioelectronic therapy for uncontrolled high blood pressure has emerged from the development of a stretchable hydrogel implant. Engineers at Penn State designed CaroFlex, a 3D-printed device that adheres to arteries using a gel-like material, completely eliminating the need for damaging surgical stitches. Traditional implants made of rigid metals and plastics fail to accommodate the natural expansion and contraction of blood vessels, leading to tissue damage and weakened connections over time. The Penn State team solved this by using conductive hydrogels that mimic biological tissue flexibility. Positioned near the carotid sinus, CaroFlex sends low-frequency electrical pulses to stimulate pressure-sensitive nerve endings, effectively regulating the body’s baroreflex system. Testing in rats demonstrated a blood pressure reduction of more than 15 percent across multiple electrical settings, with surrounding tissue showing little evidence of inflammation two weeks post-implantation. The researchers plan to optimize this technology for larger animal studies and eventual human clinical trials, hoping to provide a solution for patients who fail to respond to multiple medications.

• A breakthrough in photonic computing emerged as researchers at the University of Pennsylvania created hybrid particles that behave like both light and matter. By embedding an atomically thin semiconductor inside a nanoscale optical cavity, the team generated quasiparticles called exciton-polaritons. These particles combine the speed of photons with the interaction ability of matter, solving a major hurdle in optical computing where photons typically pass through each other without interacting. The researchers demonstrated all-optical switching at an energy scale of roughly four quadrillionths of a joule, establishing a new benchmark. This development could eventually lead to sustainable AI data centers that process information entirely in the optical domain, avoiding the massive heat and energy waste associated with electronic chips.

• A world-first humanoid robot auction has been announced by Chinese e-commerce giant JD.com. Scheduled during the annual 618 shopping festival, the event aims to attract technology companies, research institutions, and robot collectors seeking early access to next-generation machines, though the full line-up remains undisclosed. This initiative is part of a massive five-year adoption plan where JD.com intends to deploy 3 million robots, 1 million autonomous vehicles, and 100,000 drones. Additionally, JD Retail targets robot bands exceeding $1.47 billion in 2026 and aims to shorten product launch cycles by 30 percent. The company's JoyInside robotics platform, led by Dai Wenjun, is expected to connect with over 10 million terminal devices this year, already integrating major players like Unitree Robotics and Noetix Robotics.

• A new open-source AI model capable of controlling whole-body humanoid movements at 300 frames per second has been released. Chinese firm Horizon Robotics developed the 4-billion-parameter robot cerebellum model, named HoloMotion-1, to push motion intelligence beyond previous parameter scales. The system uses a Mixture-of-Experts Transformer and KV-cache to achieve real-time inference on edge devices without overwhelming computing power. Horizon Robotics trained the model using a mix of curated motion capture data and real-world video reconstructions, applying sequence-level Proximal Policy Optimization for stable learning. When deployed directly onto a Unitree G1 humanoid without extra real-world training, the model successfully enabled zero-shot transfer of complex actions like dancing, crawling, and martial arts kicks, while also smoothly tracking live human commands from VR controllers.

• Heavy industrial lifting has been mastered by the Atlas humanoid robot through reinforcement learning and large-scale simulation training. Boston Dynamics revealed that the robot can rotate its torso 180 degrees, squat to pick up a mini-fridge, and carry it while adjusting to shifting internal weight. The company trained Atlas for millions of hours in parallel on GPUs, varying factors like object weight, floor friction, and grip strength to force adaptation. Boston Dynamics also highlighted a reduced sim-to-real gap due to the robot’s simplified hardware architecture, which uses only two actuator types and symmetrical limbs. Engineers eliminated cables across joints to allow continuous rotation, enabling the robot to successfully move a fridge weighing over 100 pounds despite being trained on 50-70 pound loads.

🔊 Industry Insights & Updates

• High-speed 3D tracking of objects with varying reflectivity is now possible using a new sensor prototype created by a team at the University of Arizona. Current 3D sensors struggle with mixed-reflectivity scenes, getting confused by the transition from dull tissue to glistening fluids or from a matte brick wall to a shiny bumper. To solve this, the researchers replaced conventional frame-by-frame cameras with a neuromorphic event camera that tracks local brightness changes at ultra-high speeds. They also eliminated the need for massive deflectometry hardware by using a laser scanner to capture the entire room, allowing algorithms to separate matte and glossy surfaces and use the diffuse parts as a virtual screen. This scalable architecture could soon improve navigation for autonomous vehicles and enhance precision during robotic surgery.

• A new open-source humanoid robot kit designed for independent developers has been unveiled. Singapore-based Menlo Research created the Asimov DIY kit to make bipedal robotics more accessible outside of corporate labs. Priced at approximately $15,000, the unassembled 3.93-foot-tall platform offers over 25 degrees of freedom and ships with detailed manuals and instructional videos. The robot features a modular architecture with universal motor mounting fixtures, allowing users to easily swap or upgrade independent leg, arm, torso, and head sections. It also incorporates a parallel Revolute-Spherical-Universal ankle mechanism for better torque distribution and passive articulated toes to improve balance while reducing mechanical complexity. The software utilizes a Processor-in-the-Loop simulation approach that injects realistic communication delays and sensor noise, training an Asymmetric Actor-Critic reinforcement learning framework. This method achieves zero-shot sim-to-real transfer, enabling the robot to walk and recover from pushes without additional calibration.

• A massive deployment of over 25,000 humanoid robots has been announced for US manufacturing plants. Hyundai Motor Group will deploy the Atlas robots developed by its subsidiary Boston Dynamics across its Hyundai Motor and Kia facilities. The automaker revealed its strategy during a JPMorgan Chase session, setting a goal to build an annual production capacity of 30,000 Atlas robots by 2028. Operations are scheduled to begin at the Hyundai Motor Group Metaplant America in Georgia in 2028, followed by Kia's Georgia facility in 2029. To support this expansion, Hyundai plans to manufacture more than 300,000 actuator units annually at US factories, ensuring critical joint and muscle components are produced locally for the robotic workforce.

• A new recyclable resin for high-precision 3D printing that can be melted down and reused has been developed. Scientists at YOKOHAMA National University created the material to address the environmental impact of permanently hardened photocurable resins used in stereolithography. By utilizing anthracene, a chemical compound with reversible light-driven reactions, the team enabled the resin to return to a liquid-like state when heated. The researchers, including Shoji Maruo and Masaru Mukai, designed the resin to cure without photoinitiators through stepwise polymerization, which simplifies the composition and reduces contamination. The team successfully printed microscopic structures like a butterfly model and reused the same material for 10 print cycles, transforming a cube into a disc by heating it to 150 degrees Celsius.

Subscribe now

🧬 BioTech

• A new system of living microrobot swarms capable of assembling into custom shapes under light control has been developed by researchers. These biohybrid robots, made from green microalga and drug-carrying nanoparticles, cluster together when exposed to blue light and disperse back into a free-swimming state under red light. Scientists demonstrated that the swarms could form complex geometries like stars and arrows with high fidelity, and even split or merge while moving. To showcase medical potential, the research team created a smart bandage concept where an AI identifies a wound shape and projects light to assemble the microrobots on medical tape. Upon application, red light triggers the release of the drug-carrying robots directly into the affected tissue, achieving a 90 percent transfer rate in under two minutes.

• A new human organ-on-a-chip model has successfully recreated key features of inflammatory bowel disease, revealing unexpected biological drivers of the condition. Researchers built donor-specific colon chips using cells from IBD patients to mimic blood flow, fluid movement, and rhythmic intestinal stretching. The study demonstrated that stromal fibroblasts are not merely structural support cells but active drivers of the disease, as IBD-derived fibroblasts caused healthy epithelial cells to exhibit disease-like behavior including barrier breakdown. The research team also found that mechanical gut motion worsens inflammation and fibrosis, while immune cells moved more aggressively into epithelial tissue in the diseased chips. Published in Nature Biomedical Engineering, this platform provides a controlled human-relevant system to study complex diseases and test new therapies.

• A new initiative to develop AI-powered robotic scientific assistants has been launched by researchers at Argonne National Laboratory. The project, named RoSA or Robot Scientific Assistant for Accelerating Experimental Workflows, aims to create machines that learn laboratory procedures directly from human scientists by observing their movements and decision-making patterns. Instead of manual programming, researchers will wear sensors while performing experiments, allowing the system to capture data to train AI models. Led by senior computer scientist Nicola Ferrier and computational scientist Arvind Ramanathan, the effort is part of the U.S. Department of Energy’s Genesis Mission to double American research productivity within a decade. The team is exploring fixed-base arms, humanoid robots, and hybrid systems, which will be tested in virtual simulations before real-world deployment. The Argonne researchers hope to demonstrate a fivefold increase in task efficiency within the next year, ultimately allowing robots to handle repetitive or hazardous laboratory work safely.

  ---

💡Products/tools of the week

• A new approach to AI agent reliability launched today as Plurai introduced vibe training, a method that generates training data, validates it, and deploys custom models within minutes. The platform eliminates the need for labeled data, annotation pipelines, and prompt engineering by allowing developers to describe what their agents should and should not do. Small language models power the system, achieving sub 100ms latency, 8x lower cost compared to GPT as judge, and over 43% fewer failures. Unlike sampled evaluation methods, Plurai runs always on, and the entire system is built on published research.

• Running an entire company with AI agents is now possible with the launch of Tycoon.us. Tycoon.us introduced a platform powered by Astra, an AI CEO, who manages over ten ready-to-use AI agents including a CMO for social media and a CTO for coding. Astra also oversees Claude Code and Hermes. Users simply provide a KPI or project, such as increasing traffic tenfold or launching onboarding flows, and Astra creates a plan, assigns the appropriate agents, tracks progress, and requests approval when necessary. All agents work out of the box, eliminating the need for setup, coding, or API keys.

• Finishing the job on AI-generated apps is the focus of WeWeb, which launched its platform to help builders overcome the typical 80 percent completion limit. WeWeb enables users to generate applications with AI and then take full control using a robust no-code editor to adjust UI, data, and logic without touching complex code. Projects can be launched instantly on infrastructure that scales to millions or exported for deployment elsewhere. The platform targets teams ready to ship production apps like SaaS products, CRMs, and internal tools rather than just creating prototypes.

• OpenHuman resolved the frustration of memory that resets, the privacy concern of data in someone else's cloud, and the technical hurdle of using a terminal. It provides a local-first and privacy-first environment where the agent remembers everything and continuously gets smarter over time. Every feature is contained in one simple interface. OpenHuman is fully open source and installs with just one click.

  ---

• Genesis AI: GENE at work | Autonomous 1x speed

• Xynova: 🚀 Xynova Flex 2: The Ultimate Hybrid-Driver Dexterous Hand.

• NVIDIA: Meet the Startup: ANYbotics Revolutionizing Industrial Inspection with AI-Powered Autonomous Robots

• ORBIT Robotics: The future is here! HELIOS!

Welcome to this week's edition of Rushing Robotics! As we cross the midpoint of May 2026, the pace of breakthroughs shows no sign of slowing—from brain implants restoring vision to blind participants, to robots working full eight-hour autonomous shifts, the future isn't just arriving—it's accelerating. Grab your favorite beverage and settle in, because what follows will reshape how you think about what's possible.

Unlock truly uncensored AI—privately.

While most AI platforms store and analyze your conversations on their servers, Venice keeps everything local to your browser—meaning your private thoughts and queries stay truly private with no server-side storage or data harvesting.
Venice leverages Trusted Execution Environments (TEEs) for end-to-end encrypted processing—ensuring that even during computation, your data remains cryptographically protected from prying eyes.

Get a $10 credit when you register through my link. Upgrade to Venice Pro and apply the credit toward the most advanced models immediately. No filters, no logs—just pure creative freedom.

Claim your $10 credit & start creating!

[Disclosure: I earn a commission if you sign up via this link.]

🤯 Mind-Blowing

Prepare to have your perceptions stretched. This section delivers the kind of breakthroughs that make you pause and reread—twice. From wireless brain implants bypassing the eyes entirely to restore artificial vision, to a higher-level brain-computer interface that interprets intent rather than just movement, the line between science fiction and science fact continues to blur. Add in rice-sized optical sensors giving surgical robots a sense of touch, humanoid robots completing full shifts with unified AI, and researchers shattering a 40-year-old atom manipulation record, and you have a lineup that redefines what machines and humans can achieve together.

🔊 Industry Insights & Updates

The deeper currents driving technology forward deserve attention. Here we explore the developments reshaping industries and research paradigms—from a quantum material that regains superconductivity under crushing magnetic fields, to a geometry-based quantum swap gate achieving 99.1 percent precision across 17,000 qubit pairs. We also cover ultrathin solar cells turning everyday glass into power generators and a major deployment agreement putting thousands of wheeled humanoid robots into global manufacturing facilities by 2032. These aren't just incremental improvements—they're foundational shifts.

🧬 BioTech

Biology meets engineering at the cellular level. This week brings a Japanese laboratory where robots—not humans—now conduct medical experiments, a jellyfish-inspired soft robot breaking speed records while navigating tight anatomical spaces, and the isolation of a rare boron peroxide molecule under remarkably mild conditions that could transform pharmaceutical production and carbon capture. The automation of discovery itself is underway, and the implications for healthcare, manufacturing, and environmental science are staggering.

💡 Products/Tools of the Week

Every week brings fresh tools to supercharge your workflow, and this lineup is no exception. From Atomic Chat's fully local, agent-enabled LLM experience that runs entirely offline with no accounts required, to Innogath's ability to transform a single question into a fully cited, multi-chapter report with auto-generated diagrams—these are tools designed for serious productivity. SynthBoard.ai simulates an entire boardroom of specialized AI advisors that debate and challenge assumptions rather than politely agreeing, while KeyAPI gives developers a single REST key to access AI-ready social data from over 20 platforms. Whether you're a developer, researcher, founder, or executive, there's something here to elevate how you work.

🎥 Video Section

Seeing is believing. This week's visual showcases feature Unitree's GD01 manned transformable mecha, a novel approach to versatile humanoid manipulation through "touch dreaming," and RLDX-1—the dexterity-first foundation model designed specifically for robot hands. Watch the future unfold in motion.

We stand at an inflection point where the convergence of neuroscience, quantum physics, robotics, and biotechnology is creating possibilities that would have seemed fantastical just a decade ago. Each breakthrough in this newsletter isn't an isolated achievement—it's a building block in a larger architecture of human capability expansion. The most exciting part? We're still in the early chapters of this transformation, and the pace is only accelerating. The question is no longer whether the future will be extraordinary, but how quickly we'll adapt to live within it Stay hungry, stay futurish!

🤯 Mind-Blowing

• A successful implantation of a wireless brain implant has restored artificial vision signals in a third blind participant. Illinois Institute of Technology led the research, while Dr. Sepehr Sani performed the surgery at Rush University Medical Center. The Intracortical Visual Prosthesis bypasses the eyes entirely, using 34 wireless stimulators with 544 electrodes to send electrical pulses directly to the visual cortex. Following a recovery period, the participant will train at The Chicago Lighthouse to learn how to interpret these signals for navigation and basic tasks. Researchers will monitor the system's long-term safety and usability over the next one to three years.

• A world-first higher-level brain-computer interface has been successfully implanted in a paralyzed patient to interpret intent rather than just muscle movement. UCHealth and the University of Colorado Anschutz Medical Campus performed the surgery on Brandon Patterson, who has been paralyzed from the chest down since a 2017 accident. Unlike standard devices that target the primary motor cortex, this implant taps into higher-functioning brain areas responsible for planning and decision-making. Shortly after the procedure, Patterson experienced phantom sensory feedback, feeling his paralyzed fingers twitch. He is now undergoing motor training to translate his thoughts into digital actions like moving a cursor, while researchers study the long-term effects of this bidirectional learning.

• A rice-sized optical sensor has been developed to give surgical robots a sense of touch. Researchers at Shanghai Jiao Tong University in China created the 1.7-millimeter device that measures force, pressure, and twisting in all directions using light instead of electronics. The technology uses an optical fiber with a soft elastomer tip that deforms upon contact, changing the light pattern which is then analyzed to calculate force. This allows machines to detect unsafe contact early and adjust in real time during delicate procedures. The team demonstrated the sensor's ability to identify hidden structures like tumors in tissue models, aiming to improve safety in minimally invasive surgery.

• Full eight-hour autonomous shifts have been achieved by humanoid robots using a new unified AI system. Figure AI developed the Helix-02 technology to combine vision, touch, and whole-body control into a single learning network, eliminating the need for separate controllers. The company also introduced System 0, a learned whole-body controller trained on over 1,000 hours of human motion data that replaces 109,000 lines of hand-engineered C++ code. This advancement allows the robots to perform multi-minute tasks without resets, such as unloading a dishwasher for four continuous minutes. Figure AI previously tested these machines at BMW facilities where they completed 10-hour shifts and moved over 90,000 parts.

• A 40-year-old record for atom manipulation has been broken by researchers who moved tens of thousands of atoms in minutes. The team from the Massachusetts Institute of Technology (MIT) and Oak Ridge National Laboratory (ORNL) used an electron beam to create 40,000 quantum defects in just 40 minutes, a stark contrast to the 1980s IBM experiment that took hours to move 35 atoms. Julian Klein, an MIT researcher who conceived the project, explained that the technique uses very few electrons to avoid damaging the crystal while moving columns of chromium atoms. Frances Ross, a professor at MIT, noted that this method acts like a photocopier for atomic defects, allowing for the creation of stable structures in three dimensions at room temperature.

🔊 Industry Insights & Updates

• An unconventional quantum material has been shown to regain its zero-resistance state under crushing magnetic fields due to internal magnetic fluctuations. Researchers at the Institute of Science and Technology Austria (ISTA) found that uranium ditelluride (UTe2) exhibits a strong transverse magnetic susceptibility that functions as electron glue. While most superconductors are destroyed by high magnetic fields, UTe2 revives at fields between 40 and 70 Tesla. Valeska Zambra, the lead author and a PhD student, and Kimberly Modic, an assistant professor, achieved this by mounting samples smaller than a grain of salt on a cantilever. By mechanically oscillating the crystal, they simulated a changing magnetic field direction to measure the susceptibility, proving that this hidden magnetic property is responsible for the reentrant superconductivity.

• A breakthrough in quantum computing stability has been achieved by developing a new geometry-based quantum swap gate. Researchers at ETH Zurich created this operation to make neutral-atom qubits more reliable by exploiting the geometric phase, which depends on the overall path atoms take through a crystal of light rather than laser intensity. This method reduces errors caused by laser fluctuations that typically plague swap gates. The team demonstrated the technology on 17,000 qubit pairs, achieving 99.1 percent precision within a millisecond. They also successfully created half-swap gates that generate correlations between qubits, a necessity for running complex quantum algorithms.

• A breakthrough in solar technology has resulted in the creation of ultrathin, semi-transparent solar cells that can turn everyday glass surfaces into power generators. Researchers at Nanyang Technological University (NTU) developed the perovskite cells to be roughly 10,000 times thinner than a human hair while maintaining high efficiency. The team, led by Annalisa Bruno, used a thermal evaporation process inside a vacuum chamber to create highly uniform layers as thin as 10 nanometers without toxic solvents. The semi-transparent version achieved a 7.6 percent efficiency while allowing 41 percent of visible light to pass through, making it color-neutral and suitable for windows. Opaque versions reached efficiencies between 7 and 12 percent. The university has filed a patent through NTUitive and plans to improve durability for commercial use.

• A major agreement has been signed to deploy thousands of wheeled humanoid robots across global manufacturing facilities. UK robotics firm Humanoid partnered with German automotive supplier Schaeffler to integrate these systems into live operations, targeting a four-digit number of robots by 2032. The initial phase runs from December 2026 to June 2027 at two German sites, where the robots will handle boxes in Herzogenaurach and undergo testing in Schweinfurt. Operating under a Robot-as-a-Service model, Humanoid will provide the robots, fleet management, maintenance, and 24/7 support. Schaeffler will also become the preferred supplier for more than half of Humanoid's joint actuator demand through 2031, supplying a seven-digit number of actuators for the wheeled platforms.

Subscribe now

🧬 BioTech

• Medical experiments previously performed by human researchers are now being conducted by machines in a new Japanese laboratory. The Institute of Science Tokyo has launched the facility featuring ten robots, including the dual-armed Maholo LabDroid designed for delicate laboratory procedures. Operating without on-site human staff, the robots automatically carry out programmed cell cultivation and handle temperature-controlled equipment. This initiative directly addresses growing labor shortages and the need to minimize human error in repetitive experimental work. The university intends to expand the robotic workforce to approximately two thousand by twenty forty. The long-term objective is to automate nearly the entire research process, a vision shared by Insilico Medicine, which recently deployed a bipedal humanoid robot named Supervisor to advance autonomous drug discovery.

• A record-breaking jellyfish-inspired soft robot has been developed to perform complex biomedical tasks using external magnetic fields. Researchers led by Professor Quanliang Cao created the device, known as the Jellyfish Magnetic Soft Robot (J-MSR), which achieves a swimming speed of 14.85 body-lengths per second. By mimicking the asymmetric contraction and relaxation of real jellyfish through an optimized trapezoidal magnetic waveform, the robot moves efficiently without onboard power or bulky buoyancy systems. The team demonstrated the robot's ability to navigate tight spaces in an ex vivo pig stomach model and carry payloads like microneedles with high targeting accuracy.

• Cleaner industrial chemistry and advanced manufacturing may soon benefit from the isolation of a rare boron peroxide molecule. Researchers at MIT created the compound, known as dioxaborirane, under remarkably mild conditions at room temperature, avoiding the extreme cold or high pressure usually required for reactive oxygen structures. The molecule features a strained ring containing one boron and two oxygen atoms, functioning like a compressed spring that stores chemical energy. Lead author Chonghe Zhang emphasized that the ability to generate these compounds under mild conditions opens the door to entirely new chemical reactions. The newly discovered structure demonstrated two distinct behaviors, transferring oxygen atoms to other molecules for potential pharmaceutical production and reacting with carbon dioxide for possible capture applications.

  ---

💡Products/tools of the week

• Atomic Chat introduces this fully local, agent-enabled experience as a free and open-source application. It runs large language models entirely on device, supporting over 1,000 options like Llama and Gemma via GGUF and ONNX. The integration of TurboQuant ensures faster inference and better memory management, while the built-in agent capabilities allow the AI to automate tasks and keep persistent local memory. With no accounts required and all processing happening offline, the app provides a secure, uncapped, and customizable tool for developers and privacy advocates.

• Innogath transforms any question into a fully cited, multi-chapter report complete with auto-generated diagrams and an interactive visual canvas. The platform features branching research pages that keep insights accessible rather than lost in linear chat formats. Its AI-driven Deep Research pipeline ingests PDFs and web pages, cross-referencing thousands of trusted sources to ground all claims. Users can select between research and chat modes while the system auto-generates context-aware branches and exportable notebooks. The tool serves researchers, students, analysts, and founders who require fast, verifiable, and well-organized research outputs.

• SynthBoard.ai simulates a boardroom of specialized AI advisors called Synths to help users navigate complex strategic, hiring, product, or personal decisions. Users drop in a brief, and the system auto-assembles or lets them choose multiple persona-matched Synths that debate, challenge assumptions, and defend positions rather than politely agreeing. Powered by multi-LLM routing and persona tuning, the platform produces consensus-scored recommendations, ranked ideas, action items, and an audit trail. It also learns from real outcomes via integrations and allows users to build custom Synths. This makes it ideal for founders, executives, and teams who want fast, auditable, multi-perspective AI advice that scales like an affordable advisory board.

• KeyAPI introduces an AI-agent-first solution that gives developers a single REST key to access profiles, posts, videos, comments, analytics, and commerce signals from over 20 platforms. The system is built specifically for LLM pipelines, automation, and autonomous agents by streaming real-time and historical social data for up to 1,000 days as clean, AI-ready JSON. By removing the complexity of multiple keys and rate limits, KeyAPI allows teams to train models, enrich prompts, power agent workflows, and deploy scalable social intelligence much faster.

  ---

• Unitree Robotics: Unitree Unveils: GD01, A Manned Transformable Mecha, from $650,000

• Yaru Niu: Humanoid Touch Dream: Learning Versatile Humanoid Manipulation with Touch Dreaming

• RLWRLD - Dexterity Physical AI: RLDX-1 | The Dexterity-First Foundation Model for Robot Hands

Welcome to this week's edition of Rushing Robotics! We are living through a week that will likely appear in future textbooks. Laboratories are gaining minds of their own, the unconscious brain is proving smarter than we ever dared to imagine, robots are becoming affordable enough for garage inventors, and quantum computers are reaching into the molecular fabric of life itself, this week’s digest is packed with developments that are reshaping the future.

Unlock truly uncensored AI—privately.

While most AI platforms store and analyze your conversations on their servers, Venice keeps everything local to your browser—meaning your private thoughts and queries stay truly private with no server-side storage or data harvesting.
Venice leverages Trusted Execution Environments (TEEs) for end-to-end encrypted processing—ensuring that even during computation, your data remains cryptographically protected from prying eyes.

Get a $10 credit when you register through my link. Upgrade to Venice Pro and apply the credit toward the most advanced models immediately. No filters, no logs—just pure creative freedom.

Claim your $10 credit & start creating!

[Disclosure: I earn a commission if you sign up via this link.]

🤯 Mind-Blowing

We start with a look at Insilico Medicine's LabClaw, an intelligent operating system that is shifting drug discovery from rigid automation to true autonomy using AI agents and natural language commands. We also explore the startling discovery that our brains predict language even under anesthesia, examine Unitree's surprisingly affordable bipedal robot, witness hybrid quantum-classical supercomputers simulating massive biological molecules, and marvel at a new silicon-germanium chip capable of handling unprecedented data speeds for AI.

🔊 Industry Insights & Updates

In industry news, China powers up the LineShine exascale supercomputer built on a fully homegrown tech stack. We also cover Rice University's breakthrough in heat-resistant perovskite solar cells, Tohoku University's new design principles for nanoreactors that manage molecular "traffic," and Harvard's innovative 3D printing strategy for creating programmable artificial muscles.

🧬 BioTech

The BioTech section reveals how CRISPR-Cas12a2 is being weaponized to selectively shred cancer DNA while sparing healthy cells. We also examine a surprising link between androgen loss and accelerated brain tumor growth, and celebrate the groundbreaking confirmation that the human heart can grow new muscle cells after a heart attack, opening new doors for regenerative medicine.

💡 Products/Tools of the Week

This week's spotlight shines on four innovative tools: Innogath, an AI workspace that enables branching, verifiable research with context-aware memory; Nodalist AI, which bridges the gap between visual and intelligent thinking with its infinite canvas and multi-model debate room; Prompt2CAD, a browser-based tool that transforms text prompts into verified, downloadable CAD files through a self-correction loop; and PageLens AI, which delivers comprehensive website audits from a single URL in just five minutes.

🎥 Video Section

Finally, in our video section, we feature insights from Sequoia Capital and Nvidia's Jim Fan on the end game of robotics, a deep dive from Boston Dynamics into why Atlas moves the way it does, a look at a humanoid robot recovering from a push, and a creator who built a real-life C-3PO you can talk to.

The convergence of AI, robotics, and biotechnology is accelerating at a pace that is truly exhilarating. Every week brings us closer to solving some of humanity's most complex challenges, from curing diseases to building sustainable energy solutions. We stand on the brink of a new era where the impossible becomes routine, and the future is limited only by our imagination. Stay hungry, stay futurish!

🤯 Mind-Blowing

• A new intelligent laboratory operating system named LabClaw has been announced by Insilico Medicine to shift drug discovery from automation to autonomy. Built on a lightweight Agent-Guard architecture, the system complements the company’s LifeStar2 automated laboratory by establishing a three-pillar paradigm of AI thinking, automated execution, and human supervision. LabClaw integrates the PandaOmics AI platform, automated hardware, and laboratory information management systems using five collaborative AI agents and 28 skill modules. This allows researchers to use natural language to initiate complex workflows, breaking the rigidity of pre-programmed tasks and reducing high coordination costs. According to Lin Sha, Senior Director of IT, the system acts as a tireless co-pilot that frees scientists from repetitive tasks while keeping them at the center of decision-making through a Human-in-the-Loop mechanism.

• Real-time language comprehension and word prediction are occurring in the brain even during states of unconsciousness. A team from Baylor College of Medicine utilized Neuropixels probes for the first time in the hippocampus to record single-neuron activity in patients under general anesthesia for epilepsy surgery. Dr. Sameer Sheth and his colleagues discovered that the brain analyzes the world around it without conscious awareness, distinguishing between repetitive and unusual tones and identifying parts of speech. The researchers noted that the brain anticipates what comes next in a narrative, a form of predictive coding typically associated with being awake. Dr. Benjamin Hayden pointed out that this function resembles the predictive behavior of large language models in artificial intelligence. This activity suggests that consciousness may rely on broader coordination across brain regions rather than localized activity, fundamentally shifting how scientists understand the boundaries of cognition.

• A low-cost bipedal humanoid robot with an upper-body-only design has been introduced by Chinese robotics firm Unitree. With prices starting at 26,900 yuan ($4,290), the robot significantly lowers entry barriers in the sector. It replaces the traditional full-body structure with modular deployment options, including a fixed base or mobile chassis. The system offers flexible configurations with 5 or 7 degrees of freedom per arm, totaling up to 31 degrees of freedom. The base integrates binocular vision, a microphone array, and voice interaction, powered by dual 8-core CPUs and a vision module providing 10 TOPS of AI compute. This approach mirrors Unitree's previous success with affordable quadruped robots, aiming to foster a broad developer ecosystem.

• Hybrid quantum-classical supercomputing has reached a new scale by successfully simulating biological molecules with over 12,000 atoms. A collaborative team from Cleveland Clinic, RIKEN, and IBM developed a quantum-centric workflow where quantum hardware handled the most entangled parts of the calculation while classical supercomputers stitched the results together. They used up to 94 qubits across two processors, running 9,200 circuits over more than 100 hours to collect 1.3 billion measurement outcomes. The simulation focused on the T4-Lysozyme and Trypsin proteins, pushing far beyond earlier demonstrations of a 303-atom protein. Although the method does not yet outperform the best classical approaches, the researchers introduced improvements to both quantum algorithms and classical techniques that reduce computational cost, marking significant progress toward solving real-world chemistry problems.

• Ultra-fast signal processing for artificial intelligence and cloud infrastructure has taken a significant leap forward with the development of a record-breaking silicon-germanium chip. Researchers from the Heinz Nixdorf Institute at Paderborn University, as part of the PACE project, engineered the chip to achieve the highest combination of bandwidth and sampling rate ever demonstrated in a track-and-hold circuit. The system can handle more than 500 gigabits per second in a single channel, with multi-channel setups potentially exceeding 100 terabits per second. By using silicon-germanium technology, the team managed to increase switching speeds while reducing energy usage, addressing a key technical challenge in improving both parameters simultaneously. Maxim Weizel, a research associate on the project, noted that transceivers serve as ambassadors between analog and digital worlds, and that high speed provides a distinct competitive advantage, especially for AI handling large datasets. The development also involved partners like DESY and required high-performance computing to verify performance at extreme frequencies.

🔊 Industry Insights & Updates

• A new exascale supercomputer named LineShine has been introduced by China’s National Supercomputing Center (NSCC) in Shenzhen. The system is designed to reach 2 exaflops of performance once fully deployed, utilizing a fully homegrown stack of Chinese-designed chips, storage, and networking technologies. The development is occurring in phases, with the first phase using 100 Huawei Kunpeng servers totaling 12,800 CPU cores. The second phase will expand the system significantly to 20,480 computing nodes, each equipped with two ARMv9-based LX2 processors. NSCC director Lu Yutong stated that the project represents an effort to build an end-to-end domestic supercomputing stack, while official Li Xiaoli emphasized that the project demonstrates complete self-reliance and controllability across the entire stack.

• A new method to improve the heat resistance of perovskite solar cells has been developed by a team at Rice University. They added a two-dimensional perovskite and formamidinium chloride to the precursor solution to guide crystal growth and stabilize the black phase that absorbs sunlight. This approach bypasses the unstable yellow phase and introduces compressive strain in the crystal lattice. The modified films retained 98% of their efficiency after 1,200 hours at 194 degrees Fahrenheit under accelerated aging conditions. Researchers like Isaac Metcalf and Rabindranath Garai noted that this creates an energetically uphill route for degradation, making the material last longer.

• A new design principle for nanoreactors has been established by Tohoku University researchers by demonstrating that controlled limitation enhances catalytic performance. The findings, published in the Chemical Engineering Journal, show that precisely regulating molecular transport into confined catalytic spaces is as critical as the catalyst material itself. Instead of allowing unhindered access to active nanoparticles, the researchers used porous shells to introduce mild restrictions on reactant flow. This alignment between reactant arrival and catalyst processing prevents the saturation and starvation of active sites. Kanako Watanabe notes that this manages nanoscale traffic effectively, ensuring orderly access and continuous turnover. The resulting framework allows engineers to tailor shell structures for higher efficiency and reduced reliance on precious metals, reshaping how future nanoreactors are built.

• A new 3D printing strategy for creating programmable artificial muscles has been developed by researchers at Harvard’s SEAS and the Wyss Institute. The technique, called rotational multimaterial 3D printing, uses a spinning nozzle to extrude both active and passive materials into filaments that mimic biological muscle movement. The active material, a liquid crystal elastomer, shrinks when heated, while the passive material remains stationary, creating an internal tug-of-war that forces the strand to bend, twist, or coil. Inspired by natural filaments like grapevines and elephant trunks, this method pre-programs the curvature and twist during the printing process, eliminating the need for manual assembly. The researchers demonstrated functional prototypes such as temperature-sensitive filters and multi-object grippers, pointing to future applications in soft robotics and biomedicine.

Subscribe now

🧬 BioTech

• Wiping out malignancies without destroying healthy tissue has been advanced through the study of CRISPR-Cas12a2 by biochemists at Utah State University and collaborating institutions. The research, published in Nature, reveals that unlike Cas9 which makes single precise cuts in DNA, Cas12a2 uses guide RNA to bind complementary RNA and then shreds all DNA it encounters. Ryan Jackson and Kadin Crosby from USU, alongside Yang Liu from the University of Utah, showed that this system can selectively kill cells containing a single-point cancer mutation while sparing normal ones. The enzyme's extreme specificity means it only activates with a perfect RNA match. In mice, this targeted destruction reduced tumor volume by approximately 50 percent with a single treatment, offering a new way to eliminate pathogens and cancer cells.

• A surprising link between androgen loss and accelerated brain tumor growth has been discovered by scientists at Cleveland Clinic. They found that reducing hormones like testosterone in male preclinical models triggered the hypothalamus-pituitary-adrenal (HPA) axis, spiking stress hormones and creating an immunosuppressive environment that allowed glioblastoma to thrive. Analyzing data from over 1,300 men with glioblastoma, the researchers noted that supplemental testosterone was associated with a 38% lower risk of death. Justin Lathia, Ph.D., and his team determined that the brain’s security system tightens under androgen deficiency, keeping immune cells out. This finding challenges the assumption that androgens fuel these tumors and suggests that androgen deprivation therapy might be detrimental for glioblastoma patients.

• New muscle cell growth in the human heart following a heart attack has been documented by scientists, challenging the notion that cardiac damage is always permanent. A collaborative team from the University of Sydney, the Baird Institute, and the Royal Prince Alfred Hospital confirmed that the heart produces new muscle cells even when scarred. The researchers proved cardiomyocyte mitosis occurs in humans by analyzing living tissue samples taken from patients during bypass surgery. Dr. Robert Hume pointed out that this opens up new possibilities for recovery, as it was previously thought those areas were irreparably damaged. Although the natural regeneration is modest and not enough to fully restore heart function, the team believes it could be enhanced in the future. Professor Sean Lal added that this discovery creates a reliable model to test regenerative treatments, which is crucial given the urgent need for alternatives to the limited number of heart transplants performed each year.

  ---

💡Products/tools of the week

• Research that branches, connects, and never disappears is the promise behind Innogath, a new AI workspace from the Innogath team targeting anyone who needs fast, verifiable, and well-organized research output. By combining a Deep Research mode, an interactive visual canvas, and context-aware branching pages that inherit their parent's research memory, Innogath lets researchers, students, analysts, and founders explore complex topics layer by layer without losing earlier context.

• Thinking tools tend to be either visual or intelligent, rarely both — and that gap is exactly what the Nodalist team built Nodalist AI to close. The platform combines an infinite spatial canvas with three AI generation modes (Breakdown, Decision, and Generative), OCR-enabled file support across PDFs, spreadsheets, and images, folder bundles for multi-file analysis, and a six-model debate room that surfaces perspectives and disagreements a single model would never flag, making it a comprehensive workspace for anyone whose questions are too big for a single prompt.

• Furniture design went from a text prompt to a downloadable STEP file is the core loop that the Prompt2CAD team engineered into Prompt2CAD, a browser-based AI CAD tool aimed at strategists, researchers, founders, and anyone without traditional modeling skills. What sets it apart is a self-verification loop: the AI renders its own output, screenshots it from multiple angles, identifies geometric errors, and corrects them before the model ever reaches the user — ensuring the final export in STEP, DXF, OBJ, STL, or GLB actually matches what was asked for.

• A one-URL, five-minute audit that produces repair-ready output is the core promise the PageLens AI team built into PageLens AI, a browser-based auditing tool for founders and small teams shipping AI-built websites. The platform runs real Chromium scans backed by deterministic checks and AI-written analysis to surface issues across security, SEO, accessibility, performance, UX, tracking, and AI Search readiness — with screenshots and evidence attached to every finding.

  ---

• Sequoia Capital: Robotics’ End Game: Nvidia’s Jim Fan

• Boston Dynamics: Why does Atlas stand up like that?

• DPCcars: Humanoid Robot Gets Pushed and Instantly Recovers Like a Human

• Samuel Potozkin: I built the first real C-3PO that you can talk to (AI)