ETH Zurich Teaches A Robot Hand To Walk On Its Fingertips
0:00 / 0:39
News
ETH Zurich Teaches A Robot Hand To Walk On Its Fingertips
calendar_today Date:
schedule Duration: 0:39
database
Summary Report
Fingers as Legs: an untethered anthropomorphic hand from ETH Zurich crawls, steers and recovers from falls, then uses the same fingers to type and push objects.
- 01. RL policies trained in a hardware-calibrated simulator; untethered crawling, steering and fall recovery on rubber, granite, turf and grass
- 02. The same fingers then type keyboard commands blind and push objects to targets with overhead vision
ETH Zurich's Soft Robotics Lab has taught an anthropomorphic robot hand to walk on its fingertips. Reinforcement learning policies trained in a simulator calibrated from the real hardware handle crawling, steering and getting back up after a fall, untethered with power and compute on board, across rubber, granite, turf and grass. Once it arrives it can type keyboard commands blind and push objects to targets using an overhead camera. The paper is called Fingers as Legs.
