• HiPHI-Hand released openly on Hugging Face: 11 hours of synchronized dual-fisheye egocentric video, camera extrinsics referenced to the headset IMU, and 3D hand joint positions from optical motion capture
  • Captured by Noitom Robotics with micro-markers on the fingers and the SOL ego-vision headset, giving sub-millimeter 3D hand ground truth even where hands are occluded
  • Used to train NVIDIA EgoHand, a monocular 3D hand reconstruction model released under the OpenMDW license
  • Live at CoRL 2026, booth 1105, JW Marriott third floor, November 9–11: wear SOL and see your own hands reconstructed in 3D in real time, in a demonstration built by the Noitom Robotics research team

AUSTIN, Texas, Nov. 09, 2026 (GLOBE NEWSWIRE) — Noitom Robotics, a data company for Physical AI, today announced it will openly release HiPHI-Hand on Hugging Face, precision hand motion-capture data used to train NVIDIA EgoHand, a monocular 3D hand reconstruction model for egocentric video built on NVIDIA Isaac Capture. The two companies will also host a live demonstration at the Conference on Robot Learning (CoRL 2026) in Austin, November 9‒12, where attendees can wear the Noitom Robotics SOL ego-vision headset and watch their own hands reconstructed in 3D in real time by EgoHand running on NVIDIA RTX PRO GPUs.

A Media Snippet accompanying this announcement is available by clicking on this link.

EgoHand reconstructs 3D hand pose and shape from a single first-person camera. Released by NVIDIA under the OpenMDW license, which permits free commercial use, the model is trained on 14 million images combining synthetic data, in-the-wild video and multi-camera motion capture, and reports state-of-the-art zero-shot accuracy on public egocentric benchmarks at an inference latency of approximately 9 milliseconds per frame.

Noitom Robotics captured the HiPHI-Hand sequences for this purpose: hand‒object manipulation recorded with an optical motion-capture system using micro-markers on the fingers, synchronized with first-person video from the SOL ego-vision headset. The optical system provides sub-millimeter 3D hand ground truth even where hands are occluded in the first-person view. NVIDIA processed this capture into its SOMA-Hand representation as one of the real-world sources in EgoHand’s training corpus.

The data will be released as HiPHI-Hand on Hugging Face under the Noitom Robotics organization on Monday November 9, 2026. It contains the capture as delivered to NVIDIA: 11 hours of synchronized dual-fisheye egocentric video, camera extrinsics referenced to the headset IMU, and 3D hand joint positions from optical motion capture. HiPHI-Hand complements HiPHI, the whole-body motion and object-interaction dataset Noitom Robotics released on Hugging Face in August 2026, and adds hand-level precision to the same family of openly released data. Like HiPHI, it is released under the ModalityNet Open Research License v1.0: free for scientific research, education and evaluation, with commercial use licensed separately by Noitom Robotics.

Noitom Robotics has also integrated EgoHand into one of its ego-vision data pipelines, where it is being evaluated as a hand-annotation stage. Footage of EgoHand reconstructing hands from Sol recordings accompany this release.

“Hands are where people interact with the physical world, and they are also the hardest thing to measure in first-person video,” said Dr. Tristan Ruoli Dai, Founder and CEO of Noitom Robotics. “We captured these sequences with sub-millimeter precision so a model like EgoHand could learn from them, and we are releasing them so the community can build on the same ground truth. Volume and learnability have to scale together, and precise data is what makes large volumes of first-person video learnable.”

CoRL 2026 demonstration

At CoRL 2026, attendees will see EgoHand run in real time in a demonstration built by the Noitom Robotics research team. At booth 1105 on the third floor of the JW Marriott, November 9‒11, visitors can put on the Noitom Robotics SOL ego-vision headset, handle objects as they normally would, and watch their own hands reconstructed in 3D on screen as they move. The Noitom Robotics team integrated EgoHand into its ego-vision capture pipeline: first-person video streams from SOL to EgoHand, which reconstructs hand pose and shape frame by frame on an NVIDIA RTX GPU. Noitom Robotics researchers will also present three main-track papers at the conference: HiPHI, OmniContact and AdaPT.

Availability

About Noitom Robotics

Noitom Robotics is a data company for Physical AI. We build the World Compiler — the infrastructure layer that transforms real-world human interaction into machine-learnable representations for embodied intelligence. Our implementation is ModalityNet (modalitynet.com), a large-scale, human-centric, omnimodal data substrate built on three complementary corpora: HiPHI-MOV (High Precision Human Interaction — Motion with Object and Vision), capturing high-fidelity whole-body motion and loco-manipulation; HiPHI-OM (High Precision Human Interaction — Omni-Modality), capturing omni-modal interaction including contact, force, tactile, and object-state signals; and ITW-Ego and Ego Plus (In-The-Wild), capturing natural human behavior at scale in unconstrained real-world environments. Together they address the core bottleneck of Physical AI — not data scarcity, but learnability — by using compact, fully structured corpora to make vastly larger, weakly structured data learnable. Our mission is simple: make the physical world learnable.


Media contact

Roch Nakajima, Chief Marketing Officer, Noitom Robotics — roch@noitomrobotics.com

China media: Tan Yuxin (Evan) — evan@noitomrobotics.com

Primary Logo

Media gallery

About The Author
Noitom Robotics