Robot soccer players are more agile when controlled by reinforcement learning | Science News
HTML-код
- Опубликовано: 30 сен 2024
- The robots were more responsive when they learned to move on their own, versus being manually programmed. As input, the AIs received data including the positions and movements of the robot’s joints and, from external cameras, the positions of everything else in the game. The AIs had to output new joint positions.
Read more:
Video: Tuomas Haarnoja
Doesn't seem like a fair comparison. The RL was optimized on rewards like speed while user defined one was about accuracy. RL one looked like it could topple at any moment, its movements were not stable.
I think the comparison is pretty fair. Dynamic stability is probably more useful in a game anyway.
However, you do make a good point about why they didn't show a team of robots playing instead of 1 v 1. Perhaps it is still quite hard to make a team of robots play together.