RoboDojo
Bimanual manipulation benchmark with 42 simulated and 18 real-robot tasks and an organiser-run leaderboard.1
Details
About
- What it is
- Benchmark Inferred1
More
Classified by the Atlas from how the authors describe and distribute it.
- Built by
- 44 authors. Affiliations in paper: MMLab@HKU, UC Berkeley, THU, PKU, Stanford, MIT, UNC, Princeton, CMU, NUS, NTU, CUHK, IC, SJTU, NU, HKUST (GZ), ZJU, Yale. Leaderboard maintained by AI MMLab Club, a non-profit.2
More
Checked 2026-10-10.
- Released
- 2026-07 (arXiv v1 2026-07-05; README: paper and code released July 6, 2026)1
More
Checked 2026-10-10.
- Version
- No version tags (GitHub releases empty); paper arXiv v3 2026-07-081
More
Checked 2026-10-10.
- Last update
- README news 2026-09-16/17: fixed a single-frame get_obs discrepancy, updated code and Hugging Face data, fixed RGB channel order with RoboTwin, updated swap_T assets. Leaderboard 'Updated 2026.10.9'. Latest commit 2026-10-08.5
More
Checked 2026-10-10.
Setup
- Runs in
- Sim + real1
More
Checked 2026-10-10.
- Robot
- Two arms2
More
Checked 2026-10-10.
- Robot model
- Sim: ARX X5 bimanual platform. Real: ARX X5, Piper, Piper X (bimanual).2
More
Checked 2026-10-10.
- Setting
- Tabletop Inferred2
More
Fixed bimanual workstations with controlled workspace; scene count not stated. Checked 2026-10-10.
Scoring and access
- Scored by
- Success rate, Progress score2
More
Reported as 'score / SR%'. Sim headline = mean across the five dimensions. Real: each trial scored by three double-blind evaluators, averaged; score includes sub-step completion. Checked 2026-10-10.
- Trials
- Sim: 50 episodes per task in paper; leaderboard protocol 3 seeds (150 trials per task). Real: 10 trials per task (180 per policy).2
More
Checked 2026-10-10.
- Who runs it
- organiser-run: 'Reported scores are computed by the official evaluation system rather than self-reported by participants'; hidden verification layouts; checkpoint and code must be released for verified entries6
More
Checked 2026-10-10.
- Leaderboard
- Official leaderboard7
More
Rendered in a browser; page is JavaScript-only. Checked 2026-10-10.
- Code licence
- MIT4
More
Read the LICENSE file. Checked 2026-10-10.
- Data licence
- Apache-2.0 (Hugging Face dataset card YAML 'license: apache-2.0'; card has no other text). Sim assets hosted on ModelScope with no licence stated on the download page.8
More
Download page: https://robodojo-benchmark.com/doc/usage/install-and-download/ Checked 2026-10-10.
- Commercial use
- Allowed Inferred8
More
MIT code and Apache-2.0 data; asset licence unstated. Not legal advice. Checked 2026-10-10.
- Published at
- arXiv preprint (no journal-ref)1
More
Checked 2026-10-10.
Sources 8
- 1RoboDojo: A Unified Sim-and-Real Benchmark for Comprehensive Evaluation of Generalist Robot Manipulation PoliciesPaper · Jul 2026 · checked 10 Oct 2026
- 2RoboDojo: A Unified Sim-and-Real Benchmark for Comprehensive Evaluation of Generalist Robot Manipulation Policies (full text)Paper · Jul 2026 · checked 10 Oct 2026
- 3Semantic Scholar API recordIndex · checked 10 Oct 2026
- 4RoboDojo-Benchmark/RoboDojo on GitHub (blob)Repository · checked 10 Oct 2026
- 5RoboDojo-Benchmark/RoboDojo on GitHub (repository)Repository · checked 10 Oct 2026
- 6RoboDojo: A Unified Sim-and-Real Benchmark for Comprehensive Evaluation of Generalist Robot Manipulation PoliciesLeaderboard · checked 10 Oct 2026
- 7RoboDojo: A Unified Sim-and-Real Benchmark for Comprehensive Evaluation of Generalist Robot Manipulation PoliciesLeaderboard · checked 10 Oct 2026
- 8RoboDojo-Benchmark/RoboDojo on Hugging Face (dataset)Repository · checked 10 Oct 2026
Change history
- Created as a basic entry: identity facts checked at primary sources (phase 1 re-verification).