ETH's HumanoidPose recovers full-body robot pose from plain video, no internal sensors
ETHZurich · hf · 2026-10-21
ETH Zurich researchers (including Marco Hutter and Christian Holz) present HumanoidPose, a framework that recovers full-body 3D humanoid robot joint configurations from video alone, without any internal state measurements.
The method combines human-pose features with robot kinematics and self-trains on unlabeled videos. A single shared model handles multiple robot embodiments and inputs ranging from synthetic data and real recordings to internet and AI-generated videos, enabling robot learning from video. Code is open-sourced on GitHub.
More from Embodied
- Dream Labs exits stealth with $90M seed, ships Dream-1 world action model API at $3/hour — km · 2026-10-21
- General Intuition announces API coming soon, pitching itself as the frontier lab for acting in space and time — chris_j_paxton · 2026-10-21
- Florida debuts America's first manned electric flying police vehicle for surveillance — Polymarket · 2026-10-21
- Chris Paxton: the whole robotics stack won't run in the cloud — chris_j_paxton · 2026-10-21
- Robotics roundtable: data flywheels for real-world robot autonomy — chris_j_paxton · 2026-10-21
- Humanoid robot hands report: durability, not dexterity, gates the market at 200-1,250 hrs vs 10,000-hr target — MarwaEldiwiny · 2026-10-21