HiPHI (High-Precision Human Motion Dataset)
HiPHI (High-Precision Human Motion Dataset) is a large-scale, high-fidelity motion capture dataset released by Noitom Robotics, containing 617.5 hours of whole-body human motion and object interaction data. Designed to address the scarcity of high-precision, physically grounded data for embodied AI, it features sub-millimeter tracking accuracy from 132 performers. The dataset is theoretically guided by the FrameNet framework to maximize coverage of the human motion manifold. It serves as a critical resource for researchers and developers working on humanoid policy learning, imitation learning, and motion generation. By providing a standardized, open-access benchmark, HiPHI aims to lower barriers for academic research while offering a reference for commercial physical AI systems. The dataset is available on Hugging Face for non-commercial research, with enterprise licensing options provided through ModalityNet, making it a foundational asset for the next generation of humanoid robotics and digital human motion-prior models.
| Year | 2026 |
|---|---|
| Total hours | 618 |
| Task categories | manipulation, locomotion, loco-manipulation, human-robot-interaction |
| License | ModalityNet Open Research License v1.0 |
| Access | open |
| Maintainer | Noitom Robotics |
| Origin country | CN |
HiPHI is a comprehensive optical motion-capture dataset developed by Noitom Robotics to support the training and evaluation of humanoid robot policies. The dataset methodology focuses on high-fidelity capture of both whole-body human movement and complex object interactions, ensuring that the data is physically grounded and suitable for transfer to real-world robotic platforms. By utilizing a systematic approach to motion coverage, HiPHI provides a diverse set of behaviors that are essential for training robust locomotion and manipulation skills. The dataset is intended for use in imitation learning, motion-space analysis, and the development of motion-prior models for computer graphics and embodied intelligence. It serves as a public benchmark, allowing the research community to evaluate humanoid performance on a standardized, high-precision foundation that was previously restricted to proprietary industrial use cases.
Frequently asked questions
What is the primary purpose of the HiPHI dataset?
HiPHI is designed to provide a high-precision, large-scale foundation for training and evaluating humanoid robot policies, specifically for whole-body locomotion and manipulation.
How much data is included in the HiPHI release?
The dataset contains 617.5 hours of high-precision human motion and object interaction data captured from 132 performers.
Is the HiPHI dataset free to use?
It is free for non-commercial research under the ModalityNet Open Research License v1.0; commercial use requires a separate license from ModalityNet.
Where can I download the HiPHI dataset?
The dataset is publicly available for download on the Hugging Face platform.
What kind of precision does the HiPHI dataset offer?
The dataset provides sub-millimeter spatial marker tracking accuracy for full-body human motion and mesh-level object trajectories.