🤖 Figure’s Next Body Enters Testing
Good Morning, Roboticists!
We gave robots muscles, memories, dance moves, and a $30 million startup. All that's missing is anxiety, student debt, and a LinkedIn account.
Figure’s Next Body Enters Testing
TL;DR: Figure says its fourth-generation humanoid has reached design lock, with actuators undergoing tests and new battery packs and hands under development. Brett Adcock also reported 37.5 million human-video uploads across 112 countries for robot training. The October all-hands update connects new hardware to a larger learning pipeline, while Helix 3.0 remains a teaser without published capability evaluations. Read more →
Origin Stops Waiting For Operators
TL;DR: HIRO Industries showed Origin completing a small task autonomously at real speed, moving its stationary dual-arm platform beyond earlier teleoperated demonstrations. The workflow collects demonstrations, trains a policy and deploys it through Station. It is a concrete autonomy milestone for the Oregon startup, although the release provides no repeated-trial success rate, cycle-time benchmark or customer deployment evidence. Read more →
110 Humanoids Share One Clock
TL;DR: AGIBOT put 110 production Expedition A3 humanoids through a synchronized routine in Jinan, earning a Guinness record after five minutes and 43 seconds. Its LinkBots system coordinated formations using live robot states and UWB positioning. The useful signal is fleet consistency and orchestration at scale; the dance record itself does not establish reliable industrial manipulation or sustained service performance. Read more →
Two Robot Makers Clear The Gate
TL;DR: The FCC granted conditional approvals for Haystack Robotics' Violet AI device and T-Robotics' entire automated guided vehicle product class, adding two entries to its robotics exemptions. The October 23 notice ties continued approval to manufacturing relocation plans and updated vetting. For robot suppliers and buyers, the concrete development is a model-specific route through restrictions, with continuing conditions attached. Read more →
The Speedup Meets A Reality Check
TL;DR: An independent SuperMimic reimplementation produced better humanoid joint tracking than PPO in MuJoCo Warp, researcher Nico Bohlinger reported, but also more global-position drift and slower training under matched settings. His preliminary test lacks the official code and may miss implementation details. It adds a useful caution to the original compute-saving claims while leaving their published benchmark comparison unresolved. Read more →
Robot Muscles Attract $30 Million
TL;DR: Zurich startup Enact raised $30 million at pre-seed to develop robots that learn assembly tasks while sensing forces through their actuators. Founded by former 1X and Physical Intelligence researcher Manuel Yves Galliker, it is building motors, controls and models together. Cable insertion and small-batch assembly are initial targets; pricing, delivery dates and quantitative task results remain undisclosed. Read more →
The Alloy Lab Gets Robot Hands
TL;DR: Texas A&M received a $2 million WoodNext Foundation grant to build the Aggie AI Foundry, where robotic arms would mix, melt and test metal alloys under AI direction. Researchers aim to shorten development from months to a week or less. The announcement funds a future automated laboratory; it does not report an operating facility or demonstrated discovery speedup. Read more →
A Lighter Humanoid Gets Its Gym
TL;DR: NEURA unveiled 4NE1 Gen 4, a humanoid it says is 20 percent lighter, alongside the opening of TUM RoboGym near Munich Airport. The 2,300-square-meter facility combines physical practice, human demonstrations and simulation to teach practical tasks. Improved hands and local limb reflexes support the redesign, while serial production is being prepared without disclosed delivery dates. Read more →
Robots Learn Which Move Comes Next
TL;DR: AGIBOT released Physical Reasoning, a dataset linking robot decisions to action trajectories across changing task conditions. Examples include counting minibar bottles, remembering searched drawers and moving obstacles before grasping. Nine reasoning categories cover planning, memory and action prerequisites. The resource targets robots that choose the right next step, though measured task improvements remain undisclosed and its license restricts commercial use. Read more →
Handshakes Stop Following The Script
TL;DR: XPENG showed IRON switching between handshakes, fist bumps and high-fives as a tester changed gestures, including changing arms when needed. The company says VLA 2.0 updates visual reasoning every 0.2 seconds. That figure describes an update interval, not total response latency. The footage remains an R&D demonstration, with no disclosed success rate, repeated-trial count or customer deployment. Read more →
TOGETHER WITH US
AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.
We've helped promote over 500 Tech Brands. Will yours be the next?
Email our co-founder Mark directly at mark@aisecret.us if the button fails.