Releases: google-deepmind/mujoco_playground
Releases · google-deepmind/mujoco_playground
Release v0.1.0
[0.1.0] - 2026-01-08
- Pass through the MuJoCo Warp
(MjWarp) implementation to MJX, so that MuJoCo Playground environments can
train with MuJoCo Warp! You can pass through the implementation via the config
override
registry.load('CartpoleBalance', config_overrides={'impl': 'warp'}). - Update environments to utilize contact sensors and remove
collision.py. - Remove
mjx_env.initin favor ofmjx_env.make_datasincemake_data
now requires anMjModelargument rather than anmjx.Modelargument. - Add device to
mjx_env.make_data, fixes #174. - Update AutoResetWrapper to allow full resets on done. Fixes #179. Also
provides a means for doing curriculum learning via
state.info['AutoResetWrapper_done_count'], see #140. - Update dependencies to use
mujoco>=3.4andwarp-lang>=1.11.
Release v0.0.5
Release v0.0.4
Added
- Added ALOHA handover task (thanks to @Andrew-Luo1).
- Added Booster T1 joystick task.
booster.mp4
404879775-e8349789-be62-4847-8cb9-1721295f2746.mp4
Changed
- Fixed foot friction randomization for G1 tasks.
- Fix various bugs in
train_jax_ppo.py(thanks to @vincentzhang). - Fixed a small bug in the privileged state of the go1 joystick task.
Release v0.0.3
Bumps minimum Python version to 3.10.
Release v0.0.2
Initial release.