EgoNav: Bridging Learned Waypoints and Geometry-Aware Local Control for Robust Indoor Navigation
Industry
Source: arXiv cs.ROPublish time unverified
arXiv:2608.25642v1 Announce Type: new Abstract: Image-goal navigation using lightweight topological maps is a practical paradigm for indoor robot deployment: the map requires only geotagged images, and localization relies on visual matching rather than precise pose estimation. However, learned waypoint predictors can produce targets that violate geometric constraints or deviate from the global path.