Cross-Embodiment Robot Manipulation — Benchmarks & Datasets
Research Focus
Survey-style literature collection on benchmarks and datasets for cross-embodiment robot manipulation, with emphasis on 2024–2026 works. Context: factory/industrial manipulation, survey-style writing.
DOI List
10.48550/arXiv.1910.10897
10.1109/LRA.2020.2974707
10.48550/arXiv.2302.04659
10.48550/arXiv.2307.00595
10.48550/arXiv.2308.12952
10.48550/arXiv.2306.03310
10.48550/arXiv.2310.08864
10.48550/arXiv.2307.15818
10.48550/arXiv.2508.17449
10.48550/arXiv.2403.12945
10.48550/arXiv.2405.05941
10.48550/arXiv.2405.12213
10.48550/arXiv.2406.09246
10.48550/arXiv.2410.00425
10.48550/arXiv.2410.24164
10.48550/arXiv.2412.13877
10.48550/arXiv.2603.04356
10.48550/arXiv.2602.16710
Paper Nodes
- yu-2019-meta-world
- james-2019-rlbench
- gu-2023-maniskill2
- fang-2023-rh20t
- walke-2023-bridgedata-v2
- liu-2023-libero
- padalkar-2023-open-x-embodiment
- brohan-2023-rt-2
- li-2025-imitation-learning-survey
- khazatsky-2024-droid
- li-2024-simpler
- ghosh-2024-octo
- kim-2024-openvla
- tao-2024-maniskill3
- black-2024-pi0
- wu-2024-robomind
- nasiriany-2026-robocasa365
- zheng-2026-egoscale
Synthesis Matrix
| 論文 | 年份 | 貢獻類型 | Dataset/Benchmark | # Tasks | Cross-embodiment | 主要指標 |
|---|---|---|---|---|---|---|
| Meta-World | 2019 | Benchmark | Meta-World (50 tasks) | 50 | No | Success rate |
| RLBench | 2019 | Benchmark | RLBench (100 tasks) | 100 | No | Success rate |
| ManiSkill2 | 2023 | Benchmark | ManiSkill2 (20 envs) | ~20 | No | Success rate, SPL |
| RH20T | 2023 | Dataset | RH20T (110k+ demos) | ~140 | Partial | Skill diversity |
| BridgeData V2 | 2023 | Dataset | BridgeData V2 | ~71 | No | Success rate |
| LIBERO | 2023 | Benchmark | LIBERO (130 tasks) | 130 | No | Success rate |
| Open X-Embodiment | 2023 | Dataset | OXE (22 datasets) | 527 | Yes | Cross-robot transfer |
| RT-2 | 2023 | Model+Eval | RT-2 eval | — | Partial | Generalization score |
| IL Survey | 2025 | Survey | Multi-benchmark | — | Yes | Comprehensive review |
| DROID | 2024 | Dataset | DROID (76k trajs) | 84 | No | Task success |
| SIMPLER | 2024 | Benchmark | SIMPLER (sim eval) | — | No | Sim-to-real correlation |
| Octo | 2024 | Model+Eval | OXE (800k trajs) | — | Yes | Finetuning success |
| OpenVLA | 2024 | Model+Eval | OXE (970k demos) | — | Yes | Task success, lang grounding |
| ManiSkill3 | 2024 | Benchmark | ManiSkill3 (12 domains) | — | No | GPU sim throughput |
| π0 | 2024 | Model+Eval | Internal + OXE | — | Yes | Zero-shot & finetuned SR |
| RoboMIND | 2024 | Dataset | RoboMIND (479 tasks) | 479 | Yes | Success rate, 4 robot types |
| RoboCasa365 | 2026 | Benchmark | RoboCasa365 (household) | — | No | Multi-task, lifelong SR |
| EgoScale | 2026 | Dataset+Method | EgoScale (20,854h) | — | Yes | Success rate (+54%) |