Off-Dynamics Reinforcement Learning (ODRL): Training for Transfer with Domain Classifiers
This code is in development, please go to https://github.com/google-research/google-research/tree/master/darc for latest code, and to reproduce experiments on the paper.
Paper: https://arxiv.org/abs/2006.13916
DARC - ODRL experiments are performed under the following branches: ODRL(pointMass env), ODRL_mujoco(HalfCheetah, Reacher, Ant PyBullet and in development Mujoco experiments), ODRL_lunar_lander( Lunar Lander gym env). The branches are yet to be merged.
This work is done on the top of SAC algorithm from rlkit framework by Vitchyr https://github.com/vitchyr/rlkit.