At CoRL 2026 in Austin, Noitom Open-Releases HiPHI-Hand, the Hand-Capture Data Behind NVIDIA's EgoHand
Robot learning's big week in Austin produced a giveaway on its first day: Noitom Robotics is openly releasing 11 hours of precision hand-capture data — the dataset used to train NVIDIA's EgoHand — and demonstrating it live at the JW Marriott through Wednesday.
AUSTIN — The Conference on Robot Learning (CoRL 2026) opened in Austin on Monday, and one of its first announcements was a dataset given away. Noitom Robotics said it will openly release HiPHI-Hand on Hugging Face: precision hand motion-capture data, with sub-millimeter hand ground truth and synchronized first-person video, used to train NVIDIA EgoHand, a monocular 3D hand reconstruction model for egocentric video built on NVIDIA Isaac.
The release, per the company's announcement, holds 11 hours of synchronized dual-fisheye egocentric video, camera extrinsics referenced to the headset IMU, and 3D hand joint positions from optical motion capture — the kind of ground-truth hand data that robot-learning teams normally have to capture themselves, rig by rig.
Attendees can test the pipeline on themselves: at booth 1105 on the JW Marriott's third floor, November 9-11, visitors put on Noitom's SOL ego-vision headset, handle objects as they normally would, and watch their own hands reconstructed in 3D in real time by EgoHand running on NVIDIA RTX PRO GPUs. CoRL 2026 runs through Nov. 12 in Austin.
