EgoHTR: Egocentric 4D Demonstrations of Human Terrain Traversal
EgoHTR is a dataset of over 150,000 frames capturing 4D human motion sequences in diverse, unstructured environments. It was created using multi-modal sensors like Aria glasses and 3D scanners to bridge the gap between humanoid learning and scene reconstruction. The data allows training perceptive locomotion policies on robots like the Unitree G1 by providing terrain-aligned reference motions.