Noitom Robotics Releases HiPHI, One of the Largest High-Precision Human Motion Datasets Ever Made Public, at the World Robot Conference

Available now on Hugging Face as the first public dataset of the company’s World Compiler approach

Key facts

• Noitom Robotics has publicly released HiPHI: 617.5 hours of high-precision human motion and object interaction from 132 performers, captured at sub-millimeter precision — among the largest datasets of its kind ever made publicly available.
• Built for embodied and physical AI: policies trained on HiPHI run on a physical Unitree G1 humanoid — running, sitting, crawling, carrying a box, and pulling a suitcase — with the same foundation serving humanoid learning, digital humans, and computer graphics.
• Available now at huggingface.co/datasets/noitomrobotics/HiPHI, free for the research community; commercial licensing via modalitynet.com.
  
  

BEIJING, Aug. 19, 2026 (GLOBE NEWSWIRE) -- Noitom Robotics (NR) today released HiPHI (pronounced “hi-fi”), a high-precision optical motion-capture dataset for researchers and engineers working on humanoid learning, digital humans, and computer graphics, at the World Robot Conference in Beijing. The dataset is publicly available now on Hugging Face, and is the first the company has made public under its World Compiler approach: making the physical world learnable for AI systems.

The release totals 617.5 hours of motion, including left-right mirrored counterparts: 371.8 hours of whole-body human motion and 245.7 hours of human-object interaction in which each object’s trajectory and mesh are recorded in sync with the performer. The data was captured from 132 performers at 90 Hz with sub-millimeter marker tracking, and is organized around motion units drawn from FrameNet, a linguistic framework organizing human action semantics. It ships with its documentation in full: an academic paper — written by the NR research team — reporting the broadest measured motion coverage and the lowest per-frame physical-quality error values among the datasets compared, together with published quality metrics and a written evaluation guide. Policies trained on HiPHI run on a physical Unitree G1 humanoid: running, sitting, crawling, carrying a box, and pulling a suitcase.

“The bottleneck in physical AI is not how much data exists, but how much of it a machine can actually learn from,” said Dr. Tristan Ruoli Dai, Founder and CEO of Noitom Robotics. “HiPHI is our first answer: coverage designed before a single frame was captured, precision that survives the transfer to real robots, and the objects people touch recorded as part of the motion itself. Data like this is usually a company’s most closely guarded asset. We are making it public because embodied intelligence needs a shared foundation it can trust — and because our infrastructure keeps producing more.”

Datasets of this kind are scarce. The field has long faced a trade-off: internet-scale video lacks the precise physical states and interaction grounding robots need, while high-fidelity laboratory datasets cover only narrow slices of behavior — and high-precision capture is slow and costly to produce, so what exists is typically held as a proprietary asset. Public releases at this scale and precision are rare, and rarer still from a commercial data producer. To the company’s knowledge, HiPHI is among the largest high-precision human-motion datasets ever made publicly available.

For researchers, HiPHI provides a physically grounded foundation for humanoid tracking, imitation learning, motion generation, and motion-space analysis — a foundation that extends to motion-prior models for digital humans and computer graphics. For companies building physical AI systems, the public release also serves as a reference: full-depth access to the data the company produces continuously for commercial partners.

“Coverage, per-frame quality, and object state decide whether a humanoid can actually learn from motion data — so those are the things we designed for, measured, and published,” said Dr. Lei Han, Chief of Research and Development at Noitom Robotics. “Everything a researcher needs is in the release: standardized BVH, synchronized object trajectories, a semantic motion index, and the evaluation guide. Anyone can download it today and hold our numbers to the light. And it is a faithful sample of what our infrastructure produces — more than 100,000 hours a year — for our partners.”

HiPHI is available now on Hugging Face, free of charge for the research community under the ModalityNet Open Research License; commercial licensing is available through modalitynet.com. Noitom Robotics continues its launch activities at RO-MAN 2026 in Fukuoka (24–28 August), and plans further public releases, beginning with an omni-modality interaction corpus later this year. Upcoming releases will add support for widely used formats such as SMPL and SOMA.

About Noitom Robotics

Noitom Robotics is a data company specializing in omnimodal data for Physical AI. Through ModalityNet (modalitynet.com) — our implementation of the World Compiler — we transform real-world human motion, interaction, and behavior into learnable representations across three corpora: HiPHI-MOV (motion), HiPHI-OM (omni-modal interaction), and ITW (in-the-wild), giving embodied AI the structured, training-ready foundation that raw data alone cannot provide.

Media contacts — International: Tan Evan, evan@noitomrobotics.com.

Photos accompanying this announcement are available at:

https://www.globenewswire.com/NewsRoom/AttachmentNg/e1ac7594-23fd-4bed-b406-8fb0b683b93a 

https://www.globenewswire.com/NewsRoom/AttachmentNg/219069a0-24b0-4e2c-8554-3a400c4712b4 

https://www.globenewswire.com/NewsRoom/AttachmentNg/980ad4cb-b264-443b-b4ed-8bf26b1c7f78 

https://www.globenewswire.com/NewsRoom/AttachmentNg/559b4c52-b903-4c24-bf78-d874b5b34041

https://www.globenewswire.com/NewsRoom/AttachmentNg/69ee541c-ce6b-4fcc-af93-d2a63e4d77c5 

https://www.globenewswire.com/NewsRoom/AttachmentNg/051197b7-414b-4c7a-9a82-c9305c8d77e9

A video accompanying this announcement is available at:

https://www.globenewswire.com/NewsRoom/AttachmentNg/45ef2182-543f-41dd-9933-337431243630

HiPHI: A Large-Scale Benchmark for High-Precision Human Motion and Object-Interaction

Overview of HiPHI. A large-scale, high-fidelity dataset for humanoid learning

From FrameNet to HiPHI data construction.

Motion-relevant frames and lexical units define motion seeds, which are expanded through controlled factors such as speed, direction, amplitude, body-part involvement, and object/contact conditions.

Manual scripting versus FrameNet-guided construction.

Manual collection grows by adding scripts one by one, whereas HiPHI organizes collection around Frame-LU motion seeds and expands them through shared factors. This turns dataset growth into a structured and scalable expansion of the motion space.

Full kinematic motion-space visualization

All datasets are encoded by the same unsupervised body-motion encoder and projected into one shared t-SNE space. (a) Global projection with balanced sampled points per dataset. (b) Support-envelope comparison between HiPHI and BONES-SEED, the closest large-scale baseline; HiPHI’s support nearly encloses BONES-SEED. (c) Grid-based local-coverage statistics on the same embedding using a 55×55 grid.

Whole-body humanoid tracking performance

Whole-body humanoid tracking performance across different motion data sources. (a) Success rates under matched 3-hour and 20-hour data budgets. (b) Training failure-rate curves averaged over five independent runs, with shaded regions indicating variance. HiPHI achieves faster convergence and lower final failure rates across different data scales.

Scaling behavior of HiPHI for humanoid tracking

Increasing the amount of unmirrored HiPHI training data from 3 to 300 hours consistently reduces cross-dataset MPJPE on AMASS, BONES-SEED, Motion-X++, and LAFAN1, demonstrating continuous performance gains with increasing data scale.