Vid-HACO estimates hand contact from monocular video, lifting AP from 54.3 to 72.2
micoolcho · x · 2026-10-30
Vid-HACO, led by Botao Hao, introduces a video-to-contact model that estimates hand contact directly from monocular video — no gloves needed. On held-out HRDexDB it beats image-based HACO across the board: vertex F1 63.3 → 67.2 and AP 54.3 → 72.2. Commenters see ego-to-touch models as a big unlock for embodied AI, letting robots learn contact from egocentric human videos.
More from Embodied
- Chinese startup sells $1,000 AI laser device that shoots mosquitoes mid-flight — nordicinst · 2026-10-30
- NVIDIA open-sources Metatron v1, a unified humanoid control policy for Unitree G1 — Balance- · 2026-10-30
- Sim-trained quadruped HOUND lands Science Robotics cover with one policy — Scobleizer · 2026-10-30
- Inside HS Precision: the Shenzhen factory machining aluminum parts for leading humanoid robots — LoganGrasby · 2026-10-30
- Scoble echoes Silicon Valley robotics panel: banning Chinese robots means you're already losing — Scobleizer · 2026-10-30
- Lucen Robotics launches Laika: open modular robot kit that morphs into 5 robots for under $6k — Scobleizer · 2026-10-30