Weekly Piece of Future #149
From Microrobots to Stacked Chips and Programmable Biology
Hey there, fellow future-addicts!
Welcome to this week's edition of Rushing Robotics! This week’s theme: intelligence is getting smaller (microrobots, stacked chips), broader (robot ecosystems), and more biological (cell-size genetics, tunable proteins). Let’s dive in.
🤯 Mind-Blowing
Microrobots at near-microbe scale are now fully programmable and autonomous, stacked-chip architecture hints at the next big leap in AI hardware, and non-coding RNA just got a serious upgrade as a potential master control for cell size. Also: someone shipped a playable game they claim was built 100% with AI—an early sign of how fast “vibe coding” is compressing production cycles.
🔊 Industry Insights & Updates
China is building robot-friendly public training zones, Unitree is pushing a humanoid “app store” model, and on-device AI keeps accelerating toward data-center-class capability. At the same time, the Google–NextEra push shows the real limiter is still power—AI progress is becoming energy strategy.
🧬 BioTech
Researchers can now tune protein dosage in specific tissues across an animal’s whole life, cell-sheet tissue engineering got dramatically faster, and embryo development is becoming predictable with deep learning—opening doors to earlier detection of subtle developmental failures.
💡 Products/Tools of the Week
Superway speeds up trend discovery and forecasting from digital noise, SociaVault delivers structured social data via API for analytics and ML, Subject.so turns stories into coherent long-form video with continuity, and TextDeck generates polished PowerPoint decks from prompts in about a minute.
🎥 Video Section
Humanoids, safety concerns, biomimetic robots, and a clever gripper design—solid lineup if you want to see where embodied AI is heading.
The future feels closer because these advances stack: better chips enable better robots, better tools accelerate building, and better biology controls expand what’s possible in medicine. The next leap won’t be one headline—it’ll be the compound effect of many. Stay hungry, stay futurish!
🤯 Mind-Blowing
Microrobots were introduced at roughly the scale of microorganisms, with each unit described as nearly invisible to the human eye and measuring about 0.3 by 0.05 millimeters, placing them in the size range of bacteria and unicellular organisms. The University of Pennsylvania team built the robots and integrated computing hardware from the University of Michigan, creating what they claim are the smallest fully programmable autonomous robots to date. The robots can navigate complex paths, respond to temperature changes, and coordinate as groups, demonstrating combined mobility, sensing, and computing at extreme miniaturization.
A new discovery suggests cell size is directly controlled by a gene in the non-coding genome, offering a clearer genetic explanation for why cells don’t grow too large or stay too small. A SickKids-led research team reported that the long non-coding RNA CISTR-ACT serves as a primary regulator of cell size, shifting scientific thinking about the biological importance of non-coding DNA. The scientists said CISTR-ACT influences genes tied to cell growth, structure, and adhesion, meaning it may shape how cells develop and organize across multiple tissues rather than acting in a narrow, cell-specific way. In preclinical experiments, lowering CISTR-ACT levels increased cell size—such as enlarged red blood cells and observed brain-structure changes—while boosting CISTR-ACT produced smaller cells, reinforcing its direct role in size control. The team further explained that CISTR-ACT helps the protein FOSL2 bind to and regulate growth-related genes, especially during brain and bone marrow development, and they noted potential relevance to diseases where cell size matters, including cancer and anemia.
A major stacked-chip advance was presented by a US-led research team that says it signals a new chapter in chip manufacturing and could help set the direction for next-generation AI hardware. Stanford professor Subhasish Mitra, the project’s principal investigator, said breakthroughs like this are important on the path toward the far larger performance jumps—on the order of 1,000-fold—that future AI systems may demand. The team reports that the chip’s architecture achieves about a tenfold improvement compared with conventional 2D chips by combining an unprecedented density of vertical connections with a tightly integrated mix of memory and processing units. Senior author Tathagata Srimani, now at Carnegie Mellon University, compared the idea to elevator banks in a tall building: many high-speed “elevators” can move large numbers of “riders” (data) between floors (layers) at the same time, easing the traffic jams that hold back flat designs.
Proton mobility in metal oxides got a boost with a new MIT physical model that better predicts how protons move through many oxide materials, a step that could help develop room-temperature proton-conducting technologies as alternatives to lithium-based systems. Researchers led by MIT professor Bilge Yildiz say the model highlights why today’s proton conduction in these materials typically requires temperatures above 752°F (400°C), and points to pathways for making inorganic, scalable proton conductors that work at lower temperatures.
A fully playable game was unveiled that its creator claims was made entirely with AI. Developer Grolaf, also known as Crunch Fest, introduced Codex Mortis, describing it as the “first-ever” game created through 100% AI-driven development and “vibe coding,” with assets, programming, and background music generated by AI. Inspired by Vampire Survivors, the title is positioned as a necromantic survival bullet hell where players fight demonic hordes, collect ancient pages to progress, and unlock abilities. The project was reportedly built in three months, with ChatGPT playing a major role in shaping the artwork and helping maintain visual consistency, while animations leaned on custom shader techniques produced with Claude Code.
🔊 Industry Insights & Updates
A province-wide training network for embodied intelligent robots is being organized in Guangdong, with Shenzhen planning the country’s first city area built to be explicitly robot-friendly. The concept, branded the “1+1+N” Guangdong Embodied Intelligence Training Ground, was introduced by officials as Shenzhen unveiled the Shenzhen Embodied Intelligence Demonstration Zone during the Greater Bay AI and Robotics Industry Conference ending December 14. Guangdong’s plan centers on a main training site and management hub, then extends into the Shenzhen demonstration zone where robots can practice on public streets in everyday situations, and finally expands through multiple sector- and city-specific “N” sub-training facilities designed to collaborate with local governments and industries.
Humanoid robots got a new “app store” experience as Unitree Robotics unveiled what it says is the world’s first app store dedicated to humanoid robots, aiming to make robots as customizable as smartphones. Unitree, based in Hangzhou, presented the platform through videos shared on YouTube, X, and RedNote, showing how users can access and upgrade robot functions directly from mobile devices.
The Tiiny AI Pocket Lab was introduced as a compact device capable of running up to 120 billion-parameter LLMs entirely on-device, positioning Tiiny AI as the company bringing “data-center-level” AI capabilities to everyday users in an offline format. Tiiny AI said the device was officially launched on December 10 and framed it as a response to rising energy costs, sustainability concerns, and privacy threats linked to cloud-based AI.
A major buildout was unveiled to develop several gigawatt (GW) AI data center campuses across the US as part of an expanded partnership between NextEra Energy and Google Cloud. The companies said the new initiative extends an existing collaboration totaling about 3.2 GW of capacity that is already operating or under contract. They also pointed to recent contracts intended to add 600 megawatts (MW) of renewable energy capacity to Oklahoma’s grid to help meet rising power demand tied to large-scale computing.
🧬 BioTech
Protein levels were dialed up or down in specific tissues for an animal’s entire lifespan by a research team addressing a long-standing limitation in biology where scientists could remove proteins or switch off genes but could not finely tune dosage in chosen tissues over time. The researchers demonstrated the technique in the nematode worm C. elegans, adjusting protein amounts in the intestine and neurons while the animals continued to live, feed, and develop normally, which is essential for studying long-running processes like aging without disrupting basic biology. They argue this matters because aging and many diseases depend on continuous communication between organs, and a protein can be beneficial in one tissue yet harmful in another—effects that traditional genetic approaches often cannot separate.
A rapid tissue-engineering technique was created that produces mechanically strong, peelable cell sheets in roughly five hours using basic lab tools such as a silicone dish, spatula, and tweezers. The work comes from researchers at McMaster University, led by professor and biomedical engineering co-director Pavi Selaganapathy, and it was reported as a major speed-up over scaffold-free methods that can take up to two weeks. The team says the cell sheets may support applications including burn treatment, organ repair, and other regenerative medicine uses.
Single-cell embryo behavior was made forecastable through an MIT-built deep-learning model that predicts where cells will go and what they will do as the embryo forms. The researchers trained it on high-resolution videos of fruit fly embryos starting with roughly 5,000 cells, and the model learns from each cell’s position, its neighborhood, and whether it is folding or dividing across time. Focusing on gastrulation, which happens over about an hour with minute-by-minute rearrangements, the model reached about 90% accuracy in predicting how cells would fold, move, and reorganize, and it could also predict the exact minute specific events would occur. It’s a meaningful advance because early developmental coordination must be extremely precise, and the article links small early missteps to later serious outcomes, including abnormalities and early-onset diseases such as asthma and cancer; the team also says the model could help identify subtle dynamic signatures in live tissue that improve diagnosis or drug-screening tests.
💡Products/tools of the week
Emerging trends and shifting consumer behaviors are being surfaced with unprecedented speed through Superway, a new AI-powered trend research oracle that offers a comprehensive suite of intelligence tools. By employing its sophisticated Oracle AI, the platform analyzes vast amounts of digital noise to score sentiment and forecast the trajectory of potential market movers. The system is built around actionable AI-backed workflows, including an AI copilot that assists users in navigating complex data landscapes. With features such as SuperSense for initial discovery and SuperSeed to expand fleeting thoughts into robust concepts, Superway empowers innovation and marketing teams to spot opportunities earlier than the competition.
Real-time, structured social data from 25+ networks is extracted and delivered as JSON-ready feeds through SociaVault, a developer-first REST API built to remove the headache of managing scrapers. SociaVault pulls profiles, posts, comments, trends, and ads so teams can quickly build analytics dashboards, monitor competitors, and—critically—power AI/ML pipelines for training sentiment models, recommendation engines, trend predictors, and content classifiers. SociaVault pairs this data access with high concurrency so developers can scale collection without bottlenecks. SociaVault also supports pay-as-you-go credits and includes a free trial to help teams validate use cases fast.
Ideas are turned into cohesive storyboards and finished videos through Subject.so, an AI-powered film-creation platform built to help users produce long-form content without animation or editing skills. Subject.so uses narrative AI plus proprietary image and video models to generate plots, maintain consistent characters, and preserve scene continuity so stories feel unified from start to finish. Subject.so also produces stylized visuals and enables one-click video synthesis to move from concept to video quickly. Subject.so is positioned for creators, educators, and marketers who want visually consistent, monetizable stories on a faster production cycle.
Simple text prompts are converted into fully designed PowerPoint decks within about 60 seconds using TextDeck, an AI-powered slide builder that eliminates manual presentation design. TextDeck automatically structures content, applies professional layouts with proper visual hierarchy, and generates charts and data visualizations so slides feel polished and cohesive. TextDeck helps founders, PMs, marketers, consultants, and sales teams save time by producing presentation-ready decks without requiring design skills or lengthy production cycles. TextDeck is particularly useful for teams that need to prototype pitches quickly or generate consistent presentations on demand.
The tissue-specific protein tuning in worms is lowkey fascinating because it addresses what was basically an all-or-nothing problem in molecular biology. Being able to dial protein levels up or down in just the intestine while leaving neurons alone (or vice versa) opens up experiments that were impossible before. I worked briefly with C. elegans models and the binary nature of knockouts was always this frustrating limitation where you'd lose data because the organism couldn't compensate. The aging research implications are huge since different organs definitly need different protein balances over time.