This repository was archived by the owner on Sep 1, 2024. It is now read-only.
Tags: facebookresearch/mbrl-lib
Tags
Gymnasium migration (#177) * Migrated from gym to Gymnasium * gym==0.26.3 is still required for the dm_control and pybulle 8000 t-gym environments * Transition and TranistionBatch now support the terminated and truncated booleans instead of the single done * boolean previously used by gym * Migrated calls to env.reset() which now returns a tuple of obs, info instead of just obs * Migrated calls to env.step() which now returns a observation, reward, terminated, truncated, info * Migrated to Gymnasium render API, environments are instantiated with render_mode=None by default * DMC and PyBullet envs use the original gym wrappers to turn them into gym environments, then are wrapper by gymnasium.envs.GymV20Environment * All Mujoco envs use the DeepMind Mujoco bindings, mujoco-py is deprecated as a dependency * Custom Mujoco envs e.g. AntTruncatedObsEnv inherit from gymnasium.envs.mujoco_env.MujocoEnv, and access data through self.data instead of self.sim.data * Mujoco environment versions have been updated to v4 fromv2 e.g. Hopper-v4 * Fixed PlaNet to save model to a directory instead of a file name * Added follow-imports=skip to mypy CI test to allow for gymnasium/gym wrapper compatibility * Bumped black to version 0.23.1 in CI