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

IEEE Spectrum and Wiley have released a sponsored white paper from Noitom Robotics describing HiPHI, a large-scale dataset for humanoid robot learning. The paper is presented as an overview for robotics researchers and engineers and is available as a free download.

According to the description, HiPHI is a 617.5-hour whole-body human motion dataset captured with optical motion capture at sub-millimeter accuracy. It includes 245.7 hours of human-object interaction data with synchronized object trajectories and meshes.

The dataset's coverage is organized using FrameNet, a linguistic framework for human action. The paper also introduces a benchmark suite for measuring motion diversity and interaction grounding, and reports results from policies trained on the dataset and deployed on a physical Unitree G1 humanoid robot. The described aims include supporting reinforcement learning and sim-to-real transfer for tasks such as carrying, pushing, and pulling.

The paper addresses what it characterizes as a data gap: internet video shows diverse behavior but cannot capture precise physical states, while laboratory motion capture systems record accurate movement but usually cover only a narrow set of actions, according to the description.

Noitom Robotics is identified as the sponsor and builds ModalityNet, described as a human-centric data substrate for embodied AI. The white paper is published in partnership with IEEE Spectrum Magazine. Registration is required to access content on the hub.