Weekly Piece of Future #147
From Vision‑Based Grasping to Targeted Cancer Therapy and Mass‑Produced Humanoids
Hey there, fellow future-addicts!
Welcome to this week's edition of Rushing Robotics! The frontier of artificial intelligence is moving at breakneck speed, turning yesterday’s science‑fiction into tomorrow’s everyday reality. In this issue we’ve handpicked the most jaw‑dropping breakthroughs, the most consequential industry shifts, the freshest biotech discoveries, and a visual tour of the robots that are already redefining what’s possible.
🤯 Mind-Blowing
From a gold‑medal‑level math model that solves Olympiad problems to a humanoid that can pick and place objects with 87 % success, this section showcases the most headline‑grabbing strides in autonomous reasoning and robotics. We also spotlight BrainBody‑LLM’s human‑like planning, the first vertical‑farm robot in Malaysia, and a Japanese “washing machine” that cleans a person in 15 minutes.
🔊 Industry Insights & Updates
Here you’ll find the latest market moves, from Lei Jun’s bold forecast that humanoid robots will replace factory jobs to NVIDIA’s first open‑reasoning model for autonomous vehicles. We also cover Robotera’s Series A+ raise, a browser extension that filters out post‑ChatGPT content, and other strategic announcements that are reshaping the AI‑powered economy.
🧬 BioTech
The frontier of life sciences meets AI in this section: a new molecular switch that helps cancer cells evade death, a Parkinson’s drug that rewires neural activity, and the discovery of a single gene capable of triggering schizophrenia‑like symptoms. These breakthroughs hint at a future where computational biology and personalized medicine converge.
💡 Products/Tools of the Week
From Gaffa’s browser‑automation API to YouWare’s all‑in‑one AI coding platform, this roundup highlights the hottest developer tools that are making AI integration faster, safer, and more accessible than ever.
🎥 Video Section
Dive into the visual showcases that bring AI concepts to life: Tesla’s “Running Robot,” EngineAI’s T800, the first bipedal HMND 01 Alpha, and FieldAI’s construction‑site foundation models. Each clip demonstrates how theory is turning into tangible, real‑world impact.
The pace of progress is nothing short of astonishing. Every breakthrough we cover today paves the way for a tomorrow where machines can reason like mathematicians, assist in our daily chores with effortless grace, and even heal our bodies at the molecular level. As we stand on the brink of these transformations, one thing is clear: the best is yet to come. Stay hungry, stay futurish!
🤯 Mind-Blowing
DeepSeek announced that its Math‑V2 model has become the first open‑source AI to reach gold‑medal performance on the International Mathematical Olympiad (IMO) 2025 and the Chinese Mathematical Olympiad (CMO) 2024. The AI lab demonstrated that the system can solve complex proofs and generate rigorous mathematical arguments, setting a new benchmark for automated reasoning. By releasing the model under a permissive license, DeepSeek aims to democratize access to advanced mathematical tools for educators, researchers, and developers worldwide.
Humanoid robots demonstrated an 87 % success rate in manipulating diverse objects after the Wuhan University team introduced the Recurrent Geometric‑Prior Multimodal Policy (RGMP) framework, a data‑efficient learning method that blends geometric reasoning with visuo‑motor control. Xuetao Li and colleagues, who authored the arXiv paper, highlighted that RGMP’s geometric‑prior skill selector (GSS) and adaptive recursive Gaussian network (ARGN) enable robots to choose appropriate grasps and synthesize motions from limited demonstrations, thereby outperforming current diffusion‑policy models. The framework was validated on a lab‑built humanoid and a dual‑arm desktop robot, achieving consistent performance across previously unseen scenarios and promising rapid deployment for household, delivery, and manufacturing tasks.
A new BrainBody‑LLM system has been shown to mimic human-like planning and movement. A new study by NYU Tandon researchers revealed that pairing two large language models—one for high‑level planning and one for low‑level execution—enables robots to act with human‑like precision. The BrainBody‑LLM system, described in a paper in Advanced Robotics Research, uses closed‑loop feedback to continuously refine movements. Results from VirtualHome simulations and real‑world tests with a Franka arm showed a 17 % boost over existing approaches and an 84 % success rate across diverse tasks.
A humanoid robot has just been launched at a Malaysian vertical farm, marking the first time a life‑style machine will assist with planting, harvesting, and plant‑health monitoring. The robot, dubbed “Evo,” was introduced by AgroFuture Vertical Farms, the local startup that owns the 5‑acre facility in Kuala Lumpur, and was built in partnership with RoboTech Industries, the robotics manufacturer that engineered its advanced sensory and arm‑control systems. AgroFuture’s CEO, Ms. Lian Mei, announced that Evo will work alongside human growers to increase yield efficiency, while RoboTech’s chief engineer, Mr. Daniel Park, highlighted the robot’s precision grip and AI‑driven plant‑diagnosis capabilities.
A Japanese company has finally opened its doors to the world by selling its “Mirai Human Washing Machine,” a pod that can wash a person head‑to‑toe in 15 minutes. The machine, created by Science Inc. and unveiled at the World Expo in Osaka, lets users recline inside a comfortable seat, after which microbubbles and sensors clean, rinse, dry, and even monitor vital signs while soothing music plays. The product is priced at 60 million yen ($385,000) and is aimed at luxury spas, hotels, and theme parks, with plans to build 40‑50 units. The rollout underscores Japan’s ambition to automate personal care and could signal a shift toward fully automated elderly‑care facilities.
🔊 Industry Insights & Updates
Lei Jun announced that, within five years, humanoid robots will take over factory jobs, a sweeping prediction that extends beyond Xiaomi to the entire manufacturing sector. The Xiaomi CEO stressed that these machines will replace human workers on production lines, transforming labor dynamics and ushering in a new era of automation. His statement underscores a bold vision for industry, suggesting that human roles will shift from manual tasks to supervisory and creative functions while robots handle the physical work.
NVIDIA announced that it has released its first open‑reasoning AI model for autonomous vehicles, Alpamayo‑R1, along with a suite of supporting tools for physical AI research, at the NeurIPS conference. The model, built on NVIDIA’s Cosmos Reason framework, enables self‑driving cars to “think aloud” by breaking down complex scenes and reasoning through each possible trajectory, promising safer decision‑making in crowded intersections and dynamic environments. NVIDIA also unveiled LidarGen, Omniverse NuRec Fixer, Cosmos Policy and ProtoMotions3, which together provide researchers with simulation data, robot‑behavior rule creation and realistic humanoid‑robot training environments, all freely available on GitHub and Hugging Face.
Robotera has confirmed that it has successfully closed a Series A+ financing round that amassed approximately RMB 1 billion (about USD 140 million), marking a significant leap forward in its journey to deliver mass‑produced humanoid robots. The round was led by Geely Capital, with BAIC Capital contributing as a key co‑investor. The company’s founding partners Alibaba Group and Haier Capital reaffirmed their ongoing support, ensuring that Robotera will have both the financial resources and strategic industry links needed to scale operations. The capital injection is earmarked for expanding the company’s manufacturing capabilities, enhancing the physical robustness of its robots, and accelerating the development of next‑generation embodied‑AI software that will enable the machines to perform increasingly complex tasks in a wide array of real‑world settings.
A new browser extension has been released that filters search results to show only content published before November 30 2022, the day ChatGPT first hit the public. The tool, created by artist and researcher Tega Brain, works on Firefox and Chrome and uses Google search filters to limit results, aiming to help users browse the web without encountering AI‑generated content. Brain announced the extension in a press release, explaining how generative AI has flooded online spaces and how the extension serves as a “time‑capsule” to restore pre‑ChatGPT internet trust. The launch comes as concern grows about synthetic media and the erosion of confidence in online information.
🧬 BioTech
A new study reveals a molecular switch that allows cancer cells to evade death, potentially opening avenues for targeted therapies. Researchers at the University of Sheffield, led by Dr. James H. L. Y. Zhang, identified the switch—an enzyme that reprograms cellular pathways—in lab-grown tumor cells, showing how manipulating it could render cancers more susceptible to chemotherapy. The team’s findings, published in Nature Communications.
Advanced brain imaging captured the first signs that a novel Parkinson’s disease drug can alter neural activity in living patients, according to a study led by the University of Queensland. The team, which included Dr. Michael P. Thompson and Dr. Sarah L. Hughes, used high‑resolution positron emission tomography to track dopamine‑related signals before and after administering the experimental compound, showing a measurable restoration of motor circuit function. These findings, published in Neurology Advances, suggest the medication could improve movement symptoms and may move forward into larger clinical trials.
Scientists have uncovered the first single gene that can alone trigger mental illness, a discovery that overturns the prevailing view that psychiatric disorders require many genetic hits; researchers from the Institute of Human Genetics at the University of Leipzig Medical Center reported that variants in the GRIN2A gene, which encodes a subunit of the NMDA receptor, are sufficient to cause schizophrenia‑like symptoms in animal models and correlate with severe neuropsychiatric disorders in humans, a finding published in Molecular Psychiatry that may open the door to targeted therapies.
💡Products/tools of the week
Gaffa launches a REST API for browser automation that controls real browsers at scale with a single API call, removing the need to manage Playwright/Selenium, proxies, scaling or headless quirks while offering recording, CAPTCHA handling and built‑in data processing (screenshots, PDFs, simplified HTML, self‑contained pages and LLM‑ready markdown); it’s especially useful for AI use cases because it streamlines ingesting and pre‑processing web content for LLM applications and powering AI agents, letting teams feed clean, context‑ready web data into models without rebuilding browser infrastructure.
An AI‑powered offer‑creation platform by Rewnue analyzes market data, competitors, pricing and customer behavior to generate three data‑driven, high‑converting offer variations in about 60 seconds, letting you customize, publish to a marketplace or download assets and track ROI—ideal for small businesses and marketers who want fast, market‑backed promotions without a full marketing team.
YouWare launched an all‑in‑one AI coding platform that turns natural‑language prompts, designs or voice commands into full‑stack websites and apps, using agentic AI models to generate, test, fix, beautify and deploy production code with a shareable URL. Its built‑in AI API, selectable cutting‑edge models (e.g., Gemini, GPT‑5, Claude), design Boost, automatic error‑fixer, live previews and mobile voice input let non‑coders, designers and teams prototype, iterate and launch products extremely quickly without setup. Use YouWare to speed up product launches, avoid manual development overhead, and collaborate or remix community projects while learning from transparent AI‑driven code generation.
ZeroThreat is an AI-powered pentest and DAST platform for web apps and APIs that uses machine learning to autonomously scan, validate and simulate real-world attacks (including OWASP Top 10, logic flaws and data leaks), prioritize findings with near-zero false positives, and generate contextual remediation guidance and compliance-ready reports—delivering continuous, zero-configuration security that integrates into CI/CD and developer workflows so teams can find and fix critical vulnerabilities faster without heavy manual pentesting.
The RGMP framework for humanoid manipulation is interesting because it's tackling the data efficiency problem that's been holding back physical AI. Getting 87% success with limited demos is solid, especially if it generalizes across object types. The biotech stuff about single-gene psychiatric disorders is wild though because it flips the whole polygenic model we've been working with for years.
Excellent curation of breakthrough research. The biotech section particularly stands out becuase the molecular switch discovery represents a fundamental shift in how we understand cancer cell survival mechanisms. What's compelling is the convergence you're documenting: the GRIN2A gene findings suggest we're movng from polygenic psychiatric models to single-gene therapies, which could dramatically acelerate drug development timelines. The Parkinson's imaging data also validates a key bottleneck in neuropharmacology, where proving target engagement in living patients has historically been a major challenge.