OpenAI: Competitive Self-Play

in deep-learning •  7 years ago  (edited)

OpenAI set up competitions between 3D robots. Each robot has a goal (push the other robot, go to other side, kick the ball). The robots learn behaviors like tackling, ducking, faking, kicking and catching, and diving for the ball. Neural network of each agent was trained with PPO (Proximal Policy Optimization).

OpenAI

Authors get paid when people like you upvote their post.
If you enjoyed what you read here, create your account today and start earning FREE STEEM!
Sort Order:  

WOW