Matrix and numerical games: Match Pennies, Asymmetric Match Pennies, f(x, y) = x^2 - y^2, f(x, y) = x^2y^2 - xy.
MuJoCo games: Kick And Defend, You Shall Not Pass, Sumo Humans, Sumo Ants.
Training an RL agent with Minimax optimization and self-play.
Training an adversarial RL agent aganist the agent trained with Minimax optimization or self-play.
Retraining the victim agents against the adversarial agents.
common.py: environments related functions.
env.py: define the game environments.
logger.py: define the logger.
utils.py: logger related functions.
ppo_selfplay.py: define the selfplay related objects: training model, act model, learner, runner.
ppo_minimax.py: define the minimax play related objects: training model, act model, learner, runner.
ppo_adv.py: define the adversarial attack related objects: training model, act model, learner, runner.
selfplay_train.py: main function of training a selfplay agent.
adv_train.py: main function of training an adversarial agent.
minimax_train.py: main function of training a set of minimax agents.
zoo_utils.py: define the policy network models.
annotated_gym_compete.py, compete.py, plot_video.py, video_utils.py, video_recorder.py: generating test video for MuJoCo games.