ETPNav: Evolving Topological Planning for Vision-Language Navigation in Continuous Environments
For VLN in continuous spaces, builds a topological map online from predicted waypoints, plans on it with a cross-modal transformer and moves with an obstacle-avoiding controller; reports >10%/>20% gains on R2R-CE/RxR-CE.面向连续环境视觉语言导航：根据预测的路点在线构建拓扑地图，用跨模态 Transformer 在图上规划，再由避障控制器执行；报告在 R2R-CE、RxR-CE 上分别比此前最好方法提升 10% 和 20% 以上。
BibTeX
@misc{an2023etpnav,
title = {{ETPNav: Evolving Topological Planning for Vision-Language Navigation in Continuous Environments}},
author = {Dong An and Hanqing Wang and Wenguan Wang and Zun Wang and Yan Huang and Keji He and Liang Wang},
year = {2023},
eprint = {2304.03047},
archivePrefix = {arXiv},
primaryClass = {cs.CV},
url = {https://arxiv.org/abs/2304.03047},
}