This project implemented gail with tensorflow which has been test by gym env "Pendulum-v0".The expert data is collected by model trained by PPO,which is saved in expert_data.
You can run "train.py" to train your own model.If you choose args.algo=="ppo",you can use ppo to train your model,and if args.algo=="gail" means you can train your model by gail.
The result in Pendulum-v0 env,The model is not trained to be very convergent, but you can still see that gail imitates expert data very well.
ppo mean rewards:
gail dloss:
gail reward(env reward):
