基于上下文感知的语义地图重规划:面向对象导航
机器人学
2024-11-05 v2
摘要
通过视觉语言模型(VLMs)进行的先行探索构建的语义地图,已被证明作为无需训练的机器人应用的有效基础。然而,现有方法假设地图的准确性,不提供有效机制来根据错误地图修正决策。为此,我们引入上下文感知重规划(Context-Aware Replanning, CARe),通过置信度得分和多视角一致性估计地图不确定性,使智能体能够在无需额外标注的情况下修正因 inaccurate 地图导致的错误决策。我们通过将其与两种现代映射主干(VLMaps 和 OpenMask3D)集成,展示了所提出方法的有效性,并在对象导航任务中观察到显著的性能提升。更多细节请参见项目页面:https://care-maps.github.io/
引用
@article{arxiv.2409.04837,
title = {Context-Aware Replanning with Pre-explored Semantic Map for Object Navigation},
author = {Po-Chen Ko and Hung-Ting Su and Ching-Yuan Chen and Jia-Fong Yeh and Min Sun and Winston H. Hsu},
journal= {arXiv preprint arXiv:2409.04837},
year = {2024}
}
备注
CoRL 2024 camera ready. The first three authors contributed equally, and their order of authorship is interchangeable. Project page: https://care-maps.github.io/