Teaching A Robot Hand To Walk

Researchers from ETH Zurich turned a commercially-available robot hand into a walking machine, adding a battery and a Raspberry Pi Zero 2 W, and training a neural network via reinforcement learning to control its twenty joints. The hand can walk in a straight line, turn, recover from falls (succeeding in 21 of 25 tests), traverse fourteen different surfaces from rubber mats to gravel and grass, push light objects, and even press keyboard keys (though it can't operate one autonomously since it lacks a camera). The researchers envision this enabling a larger robot arm to detach its hand and walk to distant objects.

Source: https://hackaday.com/2026/09/19/teaching-a-robot-hand-to-walk. 8 Sync News only summarizes and links out; content copyright belongs to the authors and original sources.