Weekly Piece of Future #170
From Organ-On-A-Chip to DIY Humanoid Kits and Vibe Training
Hey there, fellow future-addicts!
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.
🧬 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.
