# mani-skill/ManiSkill

Manipulation Skill Framework, an open source GPU parallelized robotics simulator and benchmark

Repository: https://github.com/mani-skill/ManiSkill
Canonical: https://ross.abutalabs.com/products/maniskill
Homepage: https://maniskill.ai/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: reinforcement-learning, robotics, robot-manipulation, embodied-ai, robot-learning, simulation
Last push: 2026-08-04T16:38:18+00:00

## Health v2 (maintenance only)
Score: 91/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 96, release rhythm 80, longevity 100
- inputs: {"age_days": 1492, "days_push": 29, "days_rel": 134, "gap_med": 23.5, "n_releases_24m": 13}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3264, forks 535 (observed 2026-08-28T04:07:52.866693+00:00)

## What it is
ManiSkill is an open-source GPU-parallelized robotics simulation framework and benchmark built on SAPIEN, focused on manipulation skills. It provides high-throughput synthetic data collection, a task-building API, and tuned baselines for reinforcement learning, imitation learning, and vision-language-action models.

## Use cases
- train robot manipulation policies with reinforcement learning
- collect large-scale synthetic RGBD and segmentation data on GPU
- benchmark robot learning algorithms across manipulation tasks
- run sim2real transfer of policies to real robots
- build custom simulation tasks for heterogeneous parallel environments
- evaluate vision-language-action models like Octo or RDT
- simulate humanoids and mobile manipulators in parallel

## When to choose
- you need GPU-parallelized simulation for high-throughput robot learning
- you work on manipulation skills, dexterous hands, or mobile manipulation
- you want ready-made RL/IL/VLA baselines and benchmarks
- you need real2sim evaluation or sim2real deployment examples

## When to avoid
- you only need lightweight physics simulation without GPU parallelism
- your focus is locomotion or navigation rather than manipulation
- you need a non-Python workflow or non-GPU hardware

## Facets
- artifact type: framework
- maturity: active
- function: simulation, reinforcement-learning, machine-learning, benchmarking, robotics
- domain: robotics, reinforcement-learning, simulation, artificial-intelligence, machine-learning
- platform: python, cross-platform
- tags: robot-manipulation, embodied-ai, sim2real, gpu-parallel-simulation, sapien, robot-learning, benchmark, gpu, linux

## Member repositories
- mani-skill/ManiSkill (main) score 91

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:52.866693+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 2026-08-29T18:43:51.324262+00:00, confidence not recorded.
  - readme: https://github.com/mani-skill/ManiSkill (fetched 2026-08-28T04:07:52.866693+00:00, sha 7c0166f2f966)
  - homepage: https://maniskill.ai/ (fetched 2026-08-29T09:36:41.729665+00:00, sha a16d0c615537)
- Data as of 2026-08-30T08:39:29.467469+00:00.