Particle.news
Download on the App Store
4 ARTICLES
·
3d ago

ETH Zurich's ANYmal Returns Badminton Shots Autonomously

Peer-reviewed research details a vision-guided reinforcement-learning controller that coordinates locomotion with a racket arm.

Credit: ETH Zurich
Robot dog playing badminton with human being
ETH Zurich's reinforcement learning-based system enables this four-legged bot to track a shuttlecock and return shots autonomously – and without toppling over
Image

Overview

  • The system uses two cameras to track the shuttlecock, predicts its flight, and positions the robot to strike with a racket mounted on a multi-axis arm.
  • Reinforcement-learning policies trained in simulation enable precise timing and balance, letting the quadruped intercept and return shots without toppling.
  • Lab demonstrations released this week show successful rallies, including exchanges against human players, in controlled environments.
  • The work appears in Science Robotics from a team at ETH Zurich led by Marco Hutter, with the paper authored by Yuntao Ma and colleagues.
  • ANYmal is commercially used for industrial inspection, while suggestions of future sports-training applications remain speculative and the robot is reported to cost about US$150,000.