ManiSkill

ManiSkill (v1)

How to read this picture

Simulated benchmark of 4 tasks on 162 articulated objects, testing whether manipulation skills transfer to unseen object instances.1

Sources
Last checked 10 Oct 2026Basic entry18 of 23 facts checked at the sourceNext check 8 Apr 2027
Runs in
Simulation1
Checked against real robots
Not checked
Skill
Household tasks
Robot
Arm on wheels, Two arms1
Used by
2752
citations
Licence
Apache-2.03

Comparisons with real robots

Not checked1

Details

Paper limitations section states the authors had not yet conducted sim-to-real experiments. No later study found.

Details

About

What it is
Benchmark Inferred4
More

Classified by the Atlas from how the authors describe and distribute it.

Built by
UC San Diego1
More

All 9 authors at UC San Diego (Hao Su is last author). Challenge sponsored by Qualcomm (README acknowledgements).

Released
2021-074
More

arXiv v1 2021-07-30; README: initial version released 2021-07-29.

Version
Unversioned; repo renamed ManiSkill-Legacy5
More

Repo description: SAPIEN Manipulation Skill Benchmark (NeurIPS 2021 Track on Datasets and Benchmarks).

Last update
2022-08: README update announcing the ManiSkill 2022 challenge (2022-08-01); last push 2022-08-246
More

Push date from GitHub search API (https://api.github.com/search/repositories?q=ManiSkill+in:name+user:haosulab).

Status
Superseded7
More

ManiSkill3 paper: builds upon ManiSkill 1 and 2.

Setup

Runs in
Simulation1
More

SAPIEN full-physics simulator.

Robot
Arm on wheels, Two arms1
More

Moving platform (Sciurus robot body) with one or two Franka Panda arms.

Robot model
Sciurus-based mobile platform with one or two Franka Panda arms1
Setting
Whole home Inferred1
More

Household articulated objects (cabinets, swivel chairs, buckets) without room scenes; mapping to 'home' is ours.

Size
4 tasks; 162 objects in 3 categories; ~36,000 trajectories; ~1.5M frames1

Scoring and access

Scored by
Success rate1
More

Sec. 2.5.

Trials
100 evaluation trajectories per single environment; 50 per test environment over 10 test environments (paper experiments)1
More

Tables 2-3 captions.

Who runs it
The organisers run the tests6
More

README: temporary test server launched 2022-02-16; evaluate_policy.py for local use.

Leaderboard
Official leaderboard6
More

Challenge existence verified in README/paper. https://sapien.ucsd.edu/challenges/maniskill2021 refused connection (WebFetch and curl), so current content is unknown.

Code licence
Apache-2.03
Data licence
Unknown
More

README and LICENSE state no licence for assets or demonstrations. Objects are re-modelled from PartNet-Mobility, whose terms we did not check.

Sources 7

  1. 1ManiSkill: Generalizable Manipulation Skill Benchmark with Large-Scale DemonstrationsPaper · Jul 2021 · checked 10 Oct 2026
  2. 2Semantic Scholar API recordIndex · checked 10 Oct 2026
  3. 3haosulab/ManiSkill-Legacy on GitHub (main)Repository · checked 10 Oct 2026
  4. 4ManiSkill: Generalizable Manipulation Skill Benchmark with Large-Scale DemonstrationsPaper · Jul 2021 · checked 10 Oct 2026
  5. 5https://api.github.com/search/repositories?q=ManiSkill+in:name+user:haosulabRepository · checked 10 Oct 2026
  6. 6haosulab/ManiSkill-Legacy on GitHub (file README.md)Repository · checked 10 Oct 2026
  7. 7ManiSkill3: GPU Parallelized Robotics Simulation and Rendering for Generalizable Embodied AIPaper · Oct 2024 · checked 10 Oct 2026

Change history

  1. Created as a basic entry: identity facts checked at primary sources (phase 1 re-verification).