# Farama-Foundation/Metaworld

Collections of robotics environments geared towards benchmarking multi-task and meta reinforcement learning

Repository: https://github.com/Farama-Foundation/Metaworld
Canonical: https://ross.abutalabs.com/products/metaworld
Homepage: https://metaworld.farama.org/
Language: Python
License: MIT
License Family: permissive
Topics: meta-rl, multi-task, benchmark-environments, mujoco
Last push: 2026-08-10T03:57:49+00:00

## Health v2 (maintenance only)
Score: 84/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 97, release rhythm 58, longevity 100
- inputs: {"age_days": 2550, "days_push": 23, "days_rel": 66, "gap_med": 379, "n_releases_24m": 2}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1871, forks 351 (observed 2026-08-28T04:05:46.887833+00:00)

## What it is
Meta-World is an open-source benchmark of 50 simulated robotic manipulation tasks built on MuJoCo and the Gymnasium API, for evaluating multi-task and meta reinforcement learning algorithms. It provides standardized benchmark suites (MT1/MT10/MT50 and ML1/ML10/ML45) with train/test task splits for few-shot adaptation research.

## Use cases
- benchmark meta reinforcement learning algorithms
- evaluate multi-task RL policies on robotic manipulation
- test few-shot adaptation to new tasks
- compare RL algorithms on standardized continuous control tasks
- train policies across 50 manipulation tasks
- reproduce results from the Meta-World paper

## When to choose
- you need a standardized benchmark for multi-task or meta-RL research
- you want Gymnasium-compatible robotic manipulation environments
- you need train/test task splits for few-shot adaptation experiments

## When to avoid
- you need real robot hardware experiments
- you only need a single-task RL environment without benchmark structure
- you require official Windows support

## Facets
- artifact type: dataset
- maturity: active
- function: simulation, machine-learning, benchmarking, testing
- domain: reinforcement-learning, robotics, machine-learning, simulation
- platform: python
- tags: meta-rl, multi-task-learning, mujoco, gymnasium, robotic-manipulation, benchmark-environments, linux, macos

## Member repositories
- Farama-Foundation/Metaworld (main) score 84

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:46.887833+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:14:59.097234+00:00, confidence not recorded.
  - readme: https://github.com/Farama-Foundation/Metaworld (fetched 2026-08-28T04:05:46.887833+00:00, sha 37de5725b219)
  - homepage: https://metaworld.farama.org/ (fetched 2026-08-29T10:54:06.383746+00:00, sha e7e9a5f7afd6)
  - site_page: https://metaworld.farama.org/installation/installation (fetched 2026-08-29T10:54:06.393504+00:00, sha 0be376e887fb)
  - registry_pypi: https://pypi.org/pypi/metaworld/json (fetched 2026-08-29T10:54:06.396543+00:00, sha 1ca783db02cf)
- Data as of 2026-08-30T08:39:29.467469+00:00.