zanghyu / offline_rl_envs

Implementations of Gridworld, Modelwin, and Modelfail to experiment with offline RL

Enviroments in Offline RL

Implementations of Gridworld, Modelwin, and Modelfail to experiment with offline RL. The implementation is inspired by the implementation of NChainEnv found at [https://github.com/openai/gym/blob/master/gym/envs/toy_text/nchain.py].

Please see [https://arxiv.org/pdf/1604.00923.pdf] for the description of these environments.

• Grid World // Implemented in GridWorld.py. See [https://github.com/maximecb/gym-minigrid] for an existing implementation.

• ModelWin Implemented in ModelWin.py. step(): runs one step in ModelWin. reset(): be sure to call reset when done==True.

• ModelFail Implemented in ModelFail.py. step(): runs one step in ModelFail reset(): be sure to call reset when done==True.