Perceptive Behavior Foundation Model: Adapting Human Motion Priors to Robot-Centric Terrain
Perceptive BFM grounds human motion references in the robot's local terrain: synthesized terrain-conformal references train a teacher, and a student tracker adds terrain corrections only when needed.Perceptive BFM 让人类动作参考适配机器人所处的局部地形：用合成的贴合地形参考训练教师策略，学生跟踪器仅在必要时加入地形修正。
BibTeX
@misc{wang2026perceptive,
title = {{Perceptive Behavior Foundation Model: Adapting Human Motion Priors to Robot-Centric Terrain}},
author = {Zifan Wang and Yizhao Li and Teli Ma and Qiang Zhang and Yudong Fan and Hao Xu and Shuo Yang and Junwei Liang},
year = {2026},
eprint = {2606.08059},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2606.08059},
}