ETH Zurich Robot Hand Learns to Walk on Its Fingers

Researchers at ETH Zurich’s Soft Robotics Lab have built an autonomous robotic hand that can use its fingers as legs while retaining a human-like arrangement with uneven digits and an opposable thumb. Digital Trends reports that the team trained the hand with reinforcement learning in simulation, using onboard sensors, a battery, and a small computer. It can crawl across 14 surfaces, recover after falling, press keyboard keys, and push objects toward targets. In testing, the hand righted itself in 21 of 25 falls and selected the correct keyboard key 29 of 32 times. The researchers say a mobile hand could reach tight spaces without a conventional robot arm, inspect or manipulate objects independently, and then return after completing a task.

Digital Trends, Manisha Priyadarshini – A disembodied AI robot hand just learned to walk, and it looks horrifying (for a purpose) – 24th September 2026