Pytorch dqn training - Githubissues

harman097 / RoboRugby

Training environment for ML agents in a competitive robot rugby game using OpenAI's gym standard.

MIT License

4 stars 0 forks source link

Closed harman097 closed 4 years ago

harman097 commented 4 years ago

Changed framerate + moves_per_step to speed up and simplify training
Added new extension of GameEnv_Simple that uses angle/distance + trashy lidar observation space
Added Target network to Pytorch DQN (basic improvement recommended for DQN's)
Improved wall collisions so robots don't get stuck when wall sliding is turned off