ManiSkill
ManiSkillManiSkill is a GPU-parallel robot simulation and training framework built on the SAPIEN engine. It renders RGB-D and segmentation observations at tens of thousands of frames per second on one GPU, lets every parallel environment hold a different scene, and ships manipulation, mobile-manipulation and humanoid tasks with real-to-sim environments for evaluating real-world policies.
The repository moved from haosulab/ManiSkill to mani-skill/ManiSkill; the PyPI metadata still points at the old address.
Openness
2 medium confidence- license
- Apache-2.0(code and environments)+CC-BY-NC-4.0(the bundled assets)
- source
- public
- core features withheld
- no
The simulator and its environments are Apache-2.0, but the assets they load are CC BY-NC 4.0, so a company cannot use the tasks as shipped in a commercial product without replacing the assets. CC-BY-NC-4.0 covers a part the product requires and does not permit commercial use, so the license tier is noncommercial.
- https://raw.githubusercontent.com/mani-skill/ManiSkill/main/LICENSE recorded 2026-09-26
LICENSE: the Apache License, Version 2.0 text.
- https://raw.githubusercontent.com/mani-skill/ManiSkill/main/README.md recorded 2026-09-26
README license section: "All rigid body environments in ManiSkill are licensed under fully permissive licenses (e.g., Apache-2.0). The assets are licensed under CC BY-NC 4.0."
- https://ungh.cc/repos/mani-skill/ManiSkill/files/main recorded 2026-09-26
Full listing of the main branch, 1,490 paths, including per-robot asset LICENSE files and LICENSE-3RD-PARTY; no enterprise, ee or commercial path.
Adoption
2 medium confidencePyPI installs of the mani-skill package, whose metadata links to this repository under its former address.
- https://pypi.org/pypi/mani-skill/json recorded 2026-09-26
PyPI metadata for mani-skill 3.0.1: Homepage https://github.com/haosulab/ManiSkill.
- https://pypistats.org/api/packages/mani-skill/recent recorded 2026-09-26
pypistats: last_month 24,378 downloads for mani-skill.
Capability
4 medium confidenceManiSkill pairs GPU-parallel physics with rendered camera observations at very high frame rates, the same combination Genesis offers, and adds real-to-sim environments for evaluating policies trained on real robots. It has one physics engine and runs its GPU path only on NVIDIA hardware under Linux.
- https://raw.githubusercontent.com/mani-skill/ManiSkill/main/README.md recorded 2026-09-26
README: "GPU parallelized visual data collection system. On the high end you can collect RGBD + Segmentation data at 30,000+ FPS on a 4090 GPU"; "GPU parallelized heterogeneous simulation"; tasks cover "humanoids, mobile manipulators, single-arm robots"; "Real2sim environments"; powered by SAPIEN.
Verified 2026-09-26