In a serious increase to the open-source embodied AI neighborhood, motion-capture specialist Noitom Robotics simply gave humanoid robots an enormous, free library of human motion.
The firm introduced the general public launch of HiPHI on Wednesday on the World Robot Conference (WRC) 2026 in Beijing. Spanning 617.5 hours of high-precision movement seize, the benchmark, as Noitom described it, is without doubt one of the largest high-precision human movement datasets ever made public.
This is aimed instantly at coaching whole-body locomotion and manipulation insurance policies for humanoid robots.
HiPHI consists of 371.eight hours of whole-body motion and 245.7 hours of human-object interplay involving 40 bodily objects. The information was captured from 132 performers at 90 Hz with sub-millimeter precision, pairing skeletal movement with synchronized 3D object meshes and trajectories.
Unlike conventional movement seize libraries organized by session timestamps, HiPHI organizes actions utilizing FrameNet semantic indexing. This construction lets Vision-Language-Action (VLA) fashions naturally hyperlink language instructions to structured bodily actions.
Noitom mentioned insurance policies skilled on the information have already demonstrated operating, sitting, crawling, carrying a field, and pulling a suitcase on a bodily Unitree G1 humanoid.
“The bottleneck in physical AI is not how much data exists, but how much of it a machine can actually learn from,” mentioned Dr. Tristan Ruoli Dai, Founder and CEO of Noitom Robotics.
From human motion to robotic coaching
The dataset is free for non-commercial analysis on Hugging Face underneath the ModalityNet Open Research License v1.0, whereas enterprise entry is obtainable by ModalityNet.
The free launch may decrease the barrier for universities and unbiased researchers that can’t afford proprietary motion-capture datasets. More importantly, its scale, precision, and object-interaction information give researchers a shared dataset for coaching and evaluating humanoid studying programs.
HiPHI doesn’t cowl each kind of robotic manipulation. The dataset focuses totally on whole-body motion and interplay with human-scale objects slightly than high quality finger dexterity, tactile suggestions, or detailed instrument use.
Noitom says its deliberate HiPHI-OM corpus will develop into richer multimodal interplay information. The preliminary launch additionally makes use of the BVH movement format, with assist for extra body-model codecs deliberate.
An even bigger guess on open information
The launch is a part of Noitom’s broader World Compiler technique, which goals to show physical-world exercise into structured information that AI programs can be taught from.
The broader significance is {that a} business motion-data firm is placing a considerable, high-precision dataset into the general public analysis ecosystem. For universities and smaller robotics labs, that would cut back dependence on costly proprietary movement libraries and make it simpler to match coaching approaches in opposition to the identical supply materials.
“Everything a researcher needs is in the release: standardized BVH, synchronized object trajectories, a semantic motion index, and the evaluation guide,” mentioned Dr. Lei Han, chief of analysis and growth at Noitom Robotics
The extra necessary check now could be reproducibility. If unbiased researchers can use HiPHI to breed the locomotion and manipulation positive aspects Noitom has reported, the dataset may grow to be a helpful shared basis for humanoid AI analysis slightly than merely one other massive corpus.
Related News: Unitree Robotics is making ready for a Shanghai IPO as China’s humanoid robotics sector attracts rising investor consideration and pushes towards broader business deployment.
