Model-based Policy Optimization with Unsupervised Model Adaptation

This repository contains a TensorFlow implementation of AMPO.

Requirements

Install MuJoCo 1.50 at ~/.mujoco/mjpro150 and copy your license key to ~/.mujoco/mjkey.txt
pip install -r requirements.txt

Configuration files can be found in config/.

For AMPO with Wasserstein distance

python main.py --config=config.ant

For AMPO with Maximum Mean Discrepancy

python main.py --config=config.ant_mmd

Please email to rockyshen@apex.sjtu.edu.cn should you have any questions, comments or suggestions.

The code implementation is mainly modified based on the MBPO codebase.