ETH Zurich researchers make robotic hand crawl and manipulate objects

ETH Zurich researchers demonstrated a robotic hand that crawls, steers and recovers after falls using its fingers, without a separate mobility system. The prototype uses an off-the-shelf WUJI hand with onboard power and computation, allowing untethered operation. The researchers trained its movements through reinforcement learning in a simulation calibrated against physical hardware. The hand also pressed successive keyboard keys while supporting itself and pushed an object toward targets. Keyboard commands did not rely on vision; object pushing used overhead visual feedback.