Maniformer builds Physical AI data stack, reconstructs hands from egocentric video with 4.29mm error
Shruti_0810 · x · 2026-10-13
Maniformer is building one-stop data infrastructure for Physical AI, combining three sources: real-world robot data, scalable simulation data, and human demonstration data for teaching complex skills. Its MEgoVista research reconstructs 3D hand and head motion from first-person recordings, reporting 4.29mm finger articulation error against motion-capture ground truth — key to letting robots understand not just what humans do but how. The pipeline spans multi-camera input, tracking, shared virtual-plane construction, hand reconstruction, keyframe filtering, incremental rigging, and global pose recovery.
Related event: Maniformer paper hits 4.29mm hand accuracy, builds Physical AI data stack(2 posts)→
More from Embodied
- Robotaxi extends hours to 6am-11pm, adds Grok in Cybercab — yunta_tsai · 2026-10-13
- Sim-to-real skepticism: pixel-space native training beats generated simulators, argues DeWitte — chris_j_paxton · 2026-10-13
- Collov Labs launches NewEyes AI, a personal agent with eyes — Scobleizer · 2026-10-13
- Amazon now has over 1 million robot workers. What's next for human employees? — lulzxdxdxd · 2026-10-13
- Chris Paxton on IROS: dexterous, contact-rich manipulation is now robotics' frontier — chris_j_paxton · 2026-10-13
- Scoble hails non-invasive BCI from Phantom Neuro after quadriplegic user demo — Scobleizer · 2026-10-13