Xiangchen Wang

Xiangchen Wang^*, Weiye Zhu^*, Teng Wang, TianTian Geng, Zekai Zhang, Zhiyuan Qi, Jinyu Yang^†, Feng Zheng^†

A training-free runtime framework for embodied navigation that overlaps execution with observation refresh so agents move continuously instead of waiting on every inference step.

Shengli Zhou, Xiangchen Wang, Guanhua Chen^†, Feng Zheng^†

CAPruner prunes 3D scene graphs by combining fuzzy semantic relevance with spatial proximity, preserving task-critical relations for more reliable 3D spatial reasoning.

Weiye Zhu^*, Zekai Zhang^*, Xiangchen Wang^*, Hewei Pan, Teng Wang, Tiantian Geng, Rongtao Xu, Feng Zheng^†

NaVIDA is a lightweight vision-language navigation framework that uses inverse dynamics supervision to learn action-grounded visual dynamics.

Xiangchen Wang^*, Jinrui Zhang^*, Teng Wang^*, Haigang Zhang, Feng Zheng^†

LangDC compresses video tokens dynamically with lightweight language experts, allocating more visual capacity to semantically rich clips and fewer tokens to simpler segments.