ETH Zurich researchers trained a robotic hand to use its five fingers as legs for untethered locomotion, with onboard power, sensing, and computation. Reinforcement-learning policies enable the system to crawl, steer, recover from falls, and adapt its locomotion across 14