TensorFlow2.0 implementation of "Trajectory-wise Multiple Choice Learning for Dynamics Generalization in Reinforcement Learning" (NeurIPS 2020).
Since the TF1.0 code is too painful for researchers to follow, this repo upgrades the T-CML code to the more friendly TF2.0 version for further development and research by researchers.
Install required packages with below commands:
conda create -n tmcl python=3.6
pip install -r requirements.txt
Train and evaluate agents:
python -m run_scripts.run_tmcl --dataset [hopper/slim_humanoid/halfcheetah/cripple_ant] --normalize_flag