bAmpT / mgail-pytorch

Model-Based Generative Adversarial Imitation Learning

Model-Based Generative Adversarial Imitation Learning (MGAIL)

Pytorch implementation for "End-to-End Differentiable Adversarial Imitation Learning", by Nir Baram, Oron Anschel, Itai Caspi, Shie Mannor.

Dependencies

Install the required dependencies using the following command:

pip install -r requirements.txt

References

The code is based on the tensorflow implementation: https://github.com/itaicaspi/mgail

Running

Run the following command to train the Mujoco Hopper-v2 environment by imitating an expert trained with SAC or PPO:

python main.py