🤖 Returns Become The Next Robot Test
Good Morning, Roboticists!
Five eyes, six fingers, and a brain in the cloud. Frankenstein would be absolutely thrilled with this product roadmap.
Returns Become The Next Robot Test
TL;DR: Otto Group and Sereact plan to automate returns at a Polish logistics center, beginning with ten dual-arm systems and potentially expanding to 25 robots. Cortex will grasp, identify, assess and sort unfamiliar returned items under a service model. The staged project tests whether physical AI can handle changing merchandise reliably before Otto considers a wider rollout. Read more →
Show Once, Skip Retraining
TL;DR: Machenta says its simulation-trained robot model can learn a new task from one demonstration without updating its weights. The London team shows the same model across six robots and several teaching methods, including early humanoid-hand experiments. Repeated physical tests include 97 successes in 99 bottle-standing attempts, although broader task reliability and commercial deployment remain unproven. Read more →
Five Eyes Fit One Board
TL;DR: Luxonis launched OAK 4 FFC, a compact vision computer connecting five cameras in layouts chosen by robot builders. The 70-by-53-millimeter board handles synchronized capture, depth and AI processing, with scene-based calibration. Its $849 developer kit is shipping, while cameras cost extra and the barebone version arrives in 2027. The immediate benefit is simpler perception integration for custom robots. Read more →
Atlas Lines Up Its Suppliers
TL;DR: Boston Dynamics has selected a Korean-led Atlas supply chain and asked suppliers to prepare for orders covering 10,000 robots, KED Global reports. Hyundai Mobis reportedly anchors the component package, with Korean battery and camera suppliers involved. Final quantities and prices remain undecided. The development signals purchasing preparation for humanoid production, rather than a confirmed order or completed deployment. Read more →
Helix Tests A Brain Outside
TL;DR: T-Mobile and Figure are testing whether portions of Helix S1 and S2 can run closer to the network edge while preserving robot responsiveness. The work extends their existing connectivity relationship into inference. Offloading computation could ease onboard power, heat and weight constraints, but the companies have disclosed no latency results, deployment timetable or demonstrated hardware savings yet. Read more →
Underwater Robots Land Fresh Capital
TL;DR: Chasing Innovation has raised nearly RMB 100 million in Series C+ funding to develop underwater embodied AI, update products and expand internationally. Its portfolio spans industrial remotely operated vehicles and consumer water robots, with manipulation aimed at inspection, maintenance and sampling. The new round finances that expansion; existing product capabilities and factory capacity should not be mistaken for new deliveries. Read more →
Robot Welds Face The Quality Test
TL;DR: XIAOYUBOT and MCC's national steel inspection center opened a joint laboratory connecting robotic welding development with testing, traceability and workforce training. The company also cited twelve robots working at Beipao Light Steel, with August data showing higher output and lower costs. Those figures are company disclosures; the new event is the quality-validation partnership, rather than a fresh factory installation. Read more →
An Extra Thumb Reads The Task
TL;DR: EgoThumb adds a camera-guided robotic thumb to a human hand, shifting its motors to the forearm to reduce weight near the fingers. A learned policy interprets hand-and-object images to assist with holding while the wearer manipulates. Physical experiments include opening bottles and stirring tea. Hardware and software are open, but longer tasks and wider movements still reduce performance. Read more →
Dexterity Learns When To Back Off
TL;DR: Researchers trained a LEAP hand in simulation to reposition objects across varied wrist orientations, then tested it on 4,800 unseen goals over 13.3 hours. A learned safety filter reduced drops by 75 percent while retaining a 91.5 percent goal-reaching rate for the state-based policy. The results quantify a useful dexterity-safety tradeoff across four objects, rather than unrestricted manipulation. Read more →
The Hand Trades Tendons For Links
TL;DR: Torch-Hand reproduces a humanlike joint layout with 21 degrees of freedom, including five in the thumb, using linkage-based transmission instead of tendons. Its actuators sit inside the palm, with motor motion mapped directly to joint kinematics. Physical tests examine grasp coverage, force, payload and manipulation. The contribution is a hardware architecture, with no announced commercial rollout or general-purpose autonomy. 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.