

(SeaPRwire) – Egocentric video is everywhere. Learnable in-the-wild data is not. The industry obsesses over collecting viewpoints while ignoring the body behind the camera. Without body motion, video is just noise. Noitom Robotics introduced ITW-Ego Plus at CoRL 2026, aimed squarely at this blind spot. The product pairs egocentric video with full-body motion, captured by in-house Sol and Jupiter hardware. This is not another dataset. It is a statement about how data should be defined. Capturing video is the easy half. The body is the hard half.
Official claim: ITW-Ego Plus is a new category of in-the-wild human data. It pairs egocentric video with full-body motion, wrist video, external views, touch, and sound. The Body modality runs on Sol and Jupiter. Jupiter is a few small wearables, no motion-capture suit required. Industry subtext: in-the-wild data must be learnable. Video is the easy half. Body is the hard half. Without whole-body motion, video cannot teach a robot how to stand, reach, or move through a task. Noitom CEO Dr. Tristan Ruoli Dai put it plainly: ‘Scale gives you the first half; the second half is what makes scale learnable.’ Competitors sell hours of video. Noitom sells motion context.
Official facts: NEAR Lab, led by Dr. Han Lei, presented three main-track papers on Nov. 10 — HiPHI, OmniContact, and AdaPT. HiPHI is a 617.5-hour benchmark of whole-body motion and real-object interaction. OmniContact chains meta-skills via contact flow. AdaPT learns professional tennis styles from broadcast video. NEAR Lab also unveiled Unified Motion Retargeting (UMR), validated on five humanoids from 0.75 m to 1.83 m. HiPHI-Hand was released on Hugging Face under MORL v1.0. It contains 11 hours of egocentric video. Sub-millimeter 3D hand ground truth came from optical motion capture with micro-markers. NVIDIA used this capture for EgoHand. Industry subtext: Noitom is not a data dump. It is a closed loop. Capture plans what to record. Papers train on it. UMR carries motion to any robot. The 15,000 downloads of HiPHI show appetite. But the real signal is the pipeline from capture to deployment.
The supply chain for physical AI is forming. It has three layers: capture hardware, data representation, and robot retargeting. Noitom Robotics calls its infrastructure the World Compiler, implemented as ModalityNet. This layer turns human interaction into machine-learnable representations. The hardware layer is commoditizing. Sol headsets and Jupiter wearables are not the moat. The moat is the retargeting capability. UMR learns dense surface correspondence between human and robot bodies. It translates any mesh-based motion to any humanoid. That closes the loop from capture to robot. Companies selling raw video will drown in commodity data. Companies with hardware plus retargeting will define the standard. Noitom just staked its claim at CoRL 2026. The winners will own retargeting, not video hours.