Timothy Wong’s dives off the coast of Rockport and his study of robotic wheels and gears led him to develop magnetic-foot robots that clean ship hulls. The Roomba-like machines walk on steel surfaces to remove biofouling, aiming to make ship maintenance safer and more efficient.
Sanctuary AI published Part 2 of its Dexterity Explained series, arguing that dexterity is a gradient rather than one fixed target. Vacuum and parallel grippers suit simpler jobs, while highly dexterous robotic hands are needed elsewhere. The company says its own robotic hands target human-level capability, industrial reliability and performance.

Developer Brian built a VLM patrol robot on the D-Robotics RDK X5 board for the RDK Challenge Summer, combining SLAM, Ultralytics YOLO11n and InternVL2.5-1B on embedded hardware. The robot performs autonomous patrol, real-time person detection and on-device visual feature matching. D-Robotics says that stacking all three on embedded hardware was not easy, and that this is where real engineering begins.
Sharpa posted a short clip of a robot flipping a cup after the collar is removed, with nothing spilled. The demo spotlights the dexterity and grip control the company has been showing in its manipulation videos. ✌️

Sharpa walked through a two-handed task: one hand holds the Blizzard cup while the other picks up a spoon, places it, then pushes ice cream off the collar and back into the cup. The company called it a small motion with tight tolerance, part of the robot's work at a Dairy Queen store in Shanghai.

Sharpa explained how its robot adds three toppings in the right order and then blends the Blizzard. The company said long-horizon execution is the hard part: the robot must track what is done, what comes next, and whether the current step actually worked.

D-Robotics announced a livestream for D-Robotics Frontier: Uranus, on October 15 at 8:00 PM GMT+8. The session looks at how interactive world models can support long-horizon generation, multi-view simulation and robot policy evaluation.

Developer Shebin (@coderscafetech) built an AI calorie tracker on D-Robotics' RDK X5 board. Using computer vision and on-device AI, it recognizes food and shows calorie counts and nutrition info, all processed locally without a cloud connection. D-Robotics also said RDK Challenge Winter is now open for registration.
At YOLO Vision 2026, Paul Li, Head of R&D at LimX Dynamics, said vision should not be an isolated module but must work with decision-making, locomotion and manipulation inside one integrated robotic system. Lab performance is only the start: reliable robots have to handle changing lighting, spaces and tasks in real environments and keep improving through a closed loop.

Agility outlined its approach to functional safety for robots. Kevin Reese, Senior Director of Robotic Safety, says the goal is simple: keep people safe around the robot. Delivering that means designing for safety at every layer, from electrical and mechanical to industrial. The company shared a full video breakdown of how the process works.

Galbot has introduced PASSAGE, a framework that combines environmental perception, whole-body motion planning and closed-loop control, enabling humanoid robots to autonomously plan and execute flexible traversal motions based on their surroundings. The company says perceiving the environment, understanding spatial constraints and coordinating whole-body motion remain major challenges in robotics. The framework is built on 100 hours.
