AngranLi / mgail

Model-Based Generative Adversarial Imitation Learning

Model-Based Generative Adversarial Imitation Learning

Code for ICML 2017 paper "End-to-End Differentiable Adversarial Imitation Learning", by Nir Baram, Oron Anschel, Itai Caspi, Shie Mannor.

Dependencies

• Gym >= 0.8.1
• Mujoco-py >= 0.5.7
• Tensorflow >= 1.0.1

Running

Run the following command to train the Mujoco Hopper environment by imitating an expert trained with TRPO

python main.py