HandsOn Benchmark: 66-Task Dexterous Manipulation Test Shows Best Policy at 29% Success
Tianwei Luo · hf · 2026-10-20
HandsOn is a large-scale cross-embodiment simulation benchmark and platform for generalizable dexterous manipulation, evaluating generalization across robotic arms, hands, objects, and composed tasks.
Key details:
- 66 tasks (50 Core + 16 Transfer), 18,000 curated trajectories, 56 hours of human demonstrations, 3 camera views (front + two wrists)
- Core-50 success rates: OpenWAM 29.1%, FastWAM 26.4%, π0.5 24.9%, DM0.5 21.4%, LingBot-VLA 19.7%, and foundation-model-based GPT-6 (moveeef) at just 5.2%
The benchmark reveals substantial gaps in current multi-task policies and foundation-model-based control — even the best methods stay under 30% success, underscoring how far generalizable dexterous manipulation remains from being solved.
More from Embodied
- New "Embodied Olympics" Benchmark: GPT and Claude Agents Play Soccer and Ping Pong — ManlingLi_ · 2026-10-20
- SF founders go deep in Shenzhen's Huaqiangbei building BCIs and wearables — LoganGrasby · 2026-10-20
- MicroAGI trains a handyman model with tens of thousands of hours of video — isnit0 · 2026-10-20
- Uber-linked group secretly campaigns to ban robotaxis in Washington D.C. — gabriberton · 2026-10-20
- Meta admits smart glasses were recording conversations without the wake word — Resident-Fall4351 · 2026-10-20
- Wayve demos its Fiat 500e autonomous prototype on Turin's iconic Lingotto rooftop track — alexgkendall · 2026-10-20