HandsOn: A Benchmark for Dexterous Manipulation with 66 Tasks and 18,000 Human Demonstrations
清华团队构建HandsOn灵巧操作基准，66项任务全方位评测机器人双手能力
A research team from Tsinghua University, in collaboration with other institutions, has developed the HandsOn benchmark, a comprehensive evaluation framework for dexterous robotic hands. The benchmark includes 66 tasks and 18,000 human-teleoperation demonstrations, designed to assess both execution and generalization capabilities. It highlights the challenges of dexterous manipulation, including contact management and task transfer, and provides a structured approach to evaluate the performance of different strategies.
Source: Chinese robotics — Dataset and collection discovery · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
task count
66
View original evidence
HandsOn 包含 66 项人工验证任务和 18,000 条人类仿真遥操作示范，围绕工具使用、精细接触、双手协作和任务迁移展开评测。
Open source S9simulation demonstration count
18,000
View original evidence
HandsOn 包含 66 项人工验证任务和 18,000 条人类仿真遥操作示范，围绕工具使用、精细接触、双手协作和任务迁移展开评测。
Open source S9Reported success rate
29.1%
Reported success rate
<1%
Reported success rate
5.2%
View original evidence
此外，GPT-6 未经过基准数据微调，通过 move_eef 接口控制手掌与手指，在 660 次测试中，Core-50 和 Transfer-16 的成功率分别为 5.2% 和 1.9%。
Open source S13Reported success rate
1.9%
View original evidence
此外，GPT-6 未经过基准数据微调，通过 move_eef 接口控制手掌与手指，在 660 次测试中，Core-50 和 Transfer-16 的成功率分别为 5.2% 和 1.9%。
Open source S13test count
660
View original evidence
此外，GPT-6 未经过基准数据微调，通过 move_eef 接口控制手掌与手指，在 660 次测试中，Core-50 和 Transfer-16 的成功率分别为 5.2% 和 1.9%。
Open source S13task count
16
- paper: Available Artifact link Open source S16
- code: Available Artifact link Open source S16
- dataset: Available Artifact link Open source S16
- code: not_reported
- dataset: not_reported
- paper: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
HandsOn 包含 66 项人工验证任务和 18,000 条人类仿真遥操作示范，围绕工具使用、精细接触、双手协作和任务迁移展开评测。
Open source S9
结果显示，在提供训练示范的 Core-50 任务中，表现最好的受训多任务策略成功率为 29.1%。
Open source S11
面对没有目标任务训练示范的 Transfer-16，8 种受训多任务策略的任务平均成功率均低于 1%。
Open source S12
此外，GPT-6 未经过基准数据微调，通过 move_eef 接口控制手掌与手指，在 660 次测试中，Core-50 和 Transfer-16 的成功率分别为 5.2% 和 1.9%。
Open source S13
HandsOn 包含 18,000 条精选仿真遥操作示范，任务分为 Core-50 与 Transfer-16。
Open source S10
论文 arXiv：https://arxiv.org/abs/2610.26765 代码 GitHub：https://github.com/TsinghuaC3I/HandsOn 数据 Hugging Face：https://huggingface.co/datasets/Cyan666/HandsOn-Datasets 为什么灵巧手需要一套新的评测基准？
Open source S16
Source：Chinese robotics — Dataset and collection discovery · eu.36kr.com