Expriment for arcticle
IMITATION LEARNING BY REINFORCEMENT LEARNING
Based on implementation for article
Random Expert Distillation: Imitation Learning via Expert Policy Support Estimation
For environment run
pip install -r requirements
We used four models of environments provided by GYM:
'ant-bullet-medium-v0', 'halfcheetah-bullet-medium-v0', 'hopper-bullet-medium-v0', 'walker2d-bullet-medium-v0'
Expert agent is provide by d4rl-pybullet
https://github.com/takuseno/d4rl-pybullet
The experts are trained with SAC model.
We test RED and compare it with GAIL
python main.py -m algorithm=ALG/ENV
ALG is in [GAIL|RED] ENV is in [ant|halfcheetah|hopper|walker2d]
This project is based on an open source project provided by @Kaixhin