Reproduce “SurveilNav: Collaborative Object Goal Navigation with Robot and Surveillance System” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ming-Ming Yu, Qunbo Wang, Rongtao Xu, Yanghong Mei, Yirong Yang, Longteng Guo, Wenjun Wu, Jing Liu
Researchers from Beihang University and Chinese Academy of Sciences developed SurveilNav, a framework for collaborative object-goal navigation that integrates a mobile robot's egocentric perception with a static surveillance camera network. The system achieved a Success Rate weighted by Path Length (SPL) of 36.4 and a Success Rate (SR) of 71.1, outperforming state-of-the-art single-robot baselines like MCoCoNav by 6.7 SPL and 7.7 SR in large-scale, multi-floor indoor environments.
Key insights
- Fusing egocentric robot observations with third-person surveillance views provides comprehensive environmental understanding, overcoming individual perception limitations.
- Dynamically selecting and integrating relevant surveillance camera data significantly enhances navigation efficiency and target localization accuracy.
- Combining open-vocabulary object detection (Grounding DINO), instance segmentation (MobileSAM), and vision-language model (CLIP) semantic value estimation enables robust object mapping and informed waypoint selection.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.