Paper Overview
- Field: Computer Vision (CV)
- Authors: Yunsong Zhou, Xuekun Jiang, Haiqi Shen, Hangxu Liu, Chaojian Shi, Yujie Liu, Jingyuan Li, Haoran Yang, Jiafei Cao, Dongzhan Zhou, Dong Wang, Yao Mu, Chunhua Shen, Weinan Zhang
- Published: 2026-10-13
- arXiv: 2610.18872
- Calibrate: reusable calibration fixes participant geometry and sensor wear style before motion inference, decoupling mounting offsets from joint motion.
- Complete: temporal priors complete unobserved joint articulation.
- Constrain: hierarchical constraints promote measurement consistency, temporal continuity, and geometric consistency—properties that motion plausibility alone cannot guarantee.
- arXiv page: <https://arxiv.org/abs/2610.18872>
Abstract
Learning humanoid loco-manipulation from human demonstrations rests on observing how locomotion, posture, and object interaction evolve together as tasks unfold. Scaling collection across diverse environments, in turn, favors lightweight, portable capture systems. Yet, existing datasets provide only partial views of this coordination. Thus, we introduce EgoExplorer, a 100-hour vision-motion-touch observatory of human loco-manipulation built with a wearable system. With an emphasis on professional activities, its continuous recordings align egocentric video with reconstructed whole-body motion, fine-grained finger articulation, and tactile observations, accompanied by step-level descriptions.
To achieve this coverage with sparse body sensing, we develop the Calibrate-Complete-Constrain pipeline:
Relative to tracker-fitted references, the method reduces body keypoint error by 71.9% versus the strongest baseline. Paired evaluations also show improvements in hand stability and contact estimation. Preliminary studies explore multimodal generation and conditioning, as well as humanoid retargeting with whole-body control in simulation.
Links
*Auto-collected on 2026-10-15.*