Xinyan Huang

Multi-domain graph pre-training suffers from significant data redundancy; MDGMIX leverages boundary-aware subgraph mixing to achieve superior few-shot performance with improved efficiency.

Modeling foot sinkage and slip with three extra degrees of freedom unlocks more accurate bipedal robot locomotion on challenging granular terrains.

A new foot-mud interaction model enables robots to walk more efficiently on mud by morphing their feet.