search

schroederdewitt / multiagent_mujoco

Benchmark for Continuous Multi-Agent Robotic Control, based on OpenAI's Mujoco Gym environments.
Apache License 2.0
334 stars 34 forks source link

New Version/Port of `MaMuJoCo` available at Gymnasium-Robotics #23

Closed Kallinteris-Andreas closed 1 year ago

Kallinteris-Andreas commented 1 year ago

A new version of MaMuJoCo can be found here: https://robotics.farama.org/envs/MaMuJoCo/ (not associated with the original Author)

Changes:

+ Now ALL the environments work
+ Now uses the standard ‘PettingZoo.ParallelEnv‘ API.
+ Now Uses a modern version of ‘Gymnasium‘ (10.8v → 26.3v).
+ Now Uses newer MuJoCo bindings (‘mujoco-py‘ → ‘mujoco‘).
+ Now Uses newer `Gymansium.MuJoco` Environments (v2 -> v4).
+ Now includes new mapping functions (RL -> MARL, MARL -> OTHER MARL).
+ Now Has a mechanism for allowing researchers to easily create new agent factorizations.
+ Now also supports `Gymansium.MuJoCo.Pusher`.
+ Now supports observing global state.
+ Cleaned up the code base significantly.
+ Have written unit-tests and fixed a TON of bugs.
+ Have written some Documentation (There was virtually None).
+ Fixed some agent factorizations not having global observations.
- Is no longer backwards compatible with the old factorizations nor are the returns comparable.

That port is being actively maintained and well documented, and upgraded for new versions for Gymnasium/MuJoCo