HiPHI Large-Scale High-Precision Human Motion and Object Interaction Dataset Released

According to IEEE Spectrum·Robotics, HiPHI is a large-scale motion capture dataset for high-precision human motion and object interaction, aimed at filling the data gap that internet videos and existing motion capture data struggle to cover. The dataset uses FrameNet, a linguistic framework for describing human actions, to guide collection so as to systematically cover a broad range of full-body movements. The dataset also provides synchronized object trajectories and meshes, making human-object interaction data involving real tasks such as carrying, pushing, and pulling more valuable for training. The related white paper also explains that reinforcement learning policies trained on this data improve as scale increases, and can be deployed to physical humanoid robots through sim-to-real transfer. The above content was organized by AI, with information from the original source.