Abstract
Loop closures are essential for correcting odometry drift and creatingconsistent maps, especially in the context of large-scale navigation. Currentmethods using dense point clouds for accurate place recognition do not scalewell due to computationally expensive scan-to-scan comparisons. Alternativeobject-centric approaches are more efficient but often struggle withsensitivity to viewpoint variation. In this work, we introduce REGRACE, a novelapproach that addresses these challenges of scalability and perspectivedifference in re-localization by using LiDAR-based submaps. We introducerotation-invariant features for each labeled object and enhance them withneighborhood context through a graph neural network. To identify potentialrevisits, we employ a scalable bag-of-words approach, pooling one learnedglobal feature per submap. Additionally, we define a revisit with geometricalconsistency cues rather than embedding distance, allowing us to recognizefar-away loop closures. Our evaluations demonstrate that REGRACE achievessimilar results compared to state-of-the-art place recognition and registrationbaselines while being twice as fast. Code and models are publicly available.