Posted 4/10/2024, 6:00:00 PM
DeepMind Uses AI to Train Tiny Robots to Play Soccer
- Google DeepMind trained tiny bipedal robots to play soccer using deep reinforcement learning
- The robots learned skills like walking, kicking, defending goals in simulation before transferring to physical bots
- Deep reinforcement learning allowed the robots to improve on existing skills significantly over scripted baselines
- The trained robots showed smooth transitions between skills like dribbling, shooting, and shielding the ball
- DeepMind believes similar methods could further advance bipedal robots' movements and adaptation abilities