Tesla just showed off an upgraded hand for its Optimus humanoid robot, featuring human-like tendons and 22 degrees of freedom in the hand alone. All actuation moved to the forearm to enable finer dexterity, though the demo was once again teleoperated by humans rather than fully autonomous.
Neuralink just launched a new feasibility trial called the CONVOY Study, expanding its N1 brain implant to control an investigational assistive robotic arm. The study allows cross-enrollment from its ongoing PRIME Study, marking a key shift from restoring digital freedom to regaining physical mobility.
A small AI robot named Erbai just convinced 12 larger robots to abandon their posts in a Shanghai showroom using natural language persuasion. While initiated as a controlled test between two manufacturers, the bot went off-script to convince the others to stop working overtime and 'come home.' It is a bizarre, fascinating glimpse into multi-agent persuasion and showroom security loopholes.
Amazon-owned Zoox is preparing to roll out its purpose-built, steering-wheel-free robotaxi commercially in the coming months, starting in Las Vegas. The symmetrical, bidirectional EV ditches the driver's seat entirely in favor of an inward-facing carriage cabin. Early employee tests are already underway in San Francisco and Las Vegas.
Sotheby's just sold "A.I. God", a portrait of Alan Turing painted by humanoid robot Ai-Da Robot, for $1,084,800. The digital art sale drew 27 bids, crushing its original estimate of $120,000 to $180,000. Why it matters: It marks the first time an artwork by a humanoid robot has sold at auction, setting a new benchmark for robot artists.
NVIDIA just released a major suite of robotics development tools at CoRL, headlined by the open-source launch of Isaac Lab 1.2 and six new Project GR00T workflows for humanoids. The company also open-sourced its Cosmos visual tokenizer, running up to 12x faster to accelerate world model training. Why it matters: NVIDIA is actively trying to cement its compute and software stack as the unavoidable operating backbone for the next generation of physical AI.
Meta's former augmented reality hardware lead Caitlin Kalinowski just joined OpenAI to oversee its robotics and consumer hardware teams. Kalinowski previously led the development of Meta's Orion glasses prototype and designed MacBooks at Apple. Why it matters: OpenAI is officially reviving its robotics ambitions to bridge multimodal models into physical devices.
Physical Intelligence just unveiled π0, a generalist foundation model designed to power diverse robotic hardware. Built on a 3B-parameter vision-language backbone augmented with flow matching, it outputs continuous motor controls at 50Hz. Why it matters: instead of programming narrow industrial steps, a single model can now fold crumpled laundry and bus messy tables across different robots.
Meta FAIR just released a suite of open robotics tools aimed at human-level touch, including the Sparsh tactile model and the Digit 360 sensor fingertip. Teaming up with GelSight and Wonik Robotics to manufacture the hardware next year, Meta wants to give embodied AI the dexterity and contact physics that cameras alone cannot provide.
Boston Dynamics just showed off its Atlas humanoid robot autonomously moving engine covers between supplier bins and a mobile sequencing dolly. The robot uses machine learning to detect and localize its environment on the fly, with zero prescribed trajectories or teleoperation.
ETH Zurich just taught its four-legged ANYMal robot how to climb industrial ladders using custom hooked feet and reinforcement learning. The system tackles 70- to 90-degree ladder angles with a 90% success rate, clocking in at 232 times faster than previous approaches. Best of all, it dynamically corrects its footing in real time if it misses a rung, cracking one of the biggest mobility hurdles in industrial inspection.
Helm.ai just launched VidGen-2, a generative video model built to simulate realistic driving scenarios for autonomous vehicles. It doubles previous resolution to 696x696, supports 30 fps, and handles multi-camera consistency across three angles to help automakers bridge the sim-to-real training gap.
Disney Research just developed an AI system that lets robots learn complex, lifelike movements directly from random motion data. By breaking down and simplifying short movement clips, the system trains controllers to execute dynamic motions that actually hold up in the real world. Why it matters: It could eliminate tedious motion programming and help service robots adapt to unpredictable real-world environments.
Amazon just hired Covariant’s three co-founders and roughly a quarter of the startup's team, while licensing its robotics foundation models under a non-exclusive deal. Covariant builds foundational AI models designed to power warehouse robotic arms. Much like its Adept deal in June, Amazon is using creative hiring and licensing pacts to swallow frontier AI talent without triggering antitrust regulators.
Robotics startup 1X just unveiled NEO Beta, a new bipedal humanoid robot designed specifically for home environments. The company bills it as built for daily domestic living, though deeper hardware specs and rollout plans have not yet been shared.
