GLST: Global-Local Spatio-Temporal Memory Decomposition Framework for Online Chained-Goal Navigation
Navigation where goals change mid-episode without notice: memory is split into global/local spatial and long/short temporal beliefs, with goal-conditioned rewards. Tested on a Spot quadruped.研究回合中目标会无提示地改变的导航：把记忆分解为全局/局部空间与长/短时序信念，并配合目标条件奖励。在 Spot 四足机器人上实机验证。
BibTeX
@misc{liang2026glst,
title = {{GLST: Global-Local Spatio-Temporal Memory Decomposition Framework for Online Chained-Goal Navigation}},
author = {Yousef Liang and Ryan Ardalan and Yarden Goraly and Chenhao Zhu and Yue Yin and Jonas Frey and Claire Tomlin and Marco Pavone and Shankar Sastry},
year = {2026},
eprint = {2610.17927},
archivePrefix = {arXiv},
primaryClass = {cs.RO},
url = {https://arxiv.org/abs/2610.17927},
}