All Tools
M
AgentsFreeOpen Source
META-WORLD
Multi-task robotic manipulation benchmark for meta-RL
MIT
ABOUT
Meta-RL papers need many related manipulation tasks with one robot, but assembling that suite by hand is slow and inconsistent. Meta-World ships dozens of Gymnasium tabletop tasks on a shared Sawyer-style morphology so researchers can measure few-shot adaptation instead of building envs.
INSTALL
pip install metaworldINTEGRATION GUIDE
1. Benchmark a multi-task SAC agent across Meta-World ML1 or ML10 task splits
2. Evaluate few-shot meta-RL adaptation from training tasks to held-out MT50 skills
3. Collect multi-task robotic demonstrations for imitation or offline RL research
4. Compare single-task PPO against a shared multi-task policy on the same robot
TAGS
pythonmujocoroboticsmeta-rlmulti-taskgymnasiumfarama