LEADER: Learning Reliable Local-to-Global Correspondences for LiDAR Relocalization

2026-04-13

AI Summary

This work addresses the vulnerability of existing learning-based LiDAR relocalization methods to noise and outliers in complex environments, where all predicted points are treated equally regardless of their reliability. To overcome this limitation, the authors propose the LEADER framework, which incorporates a robust projection-based geometric encoder to extract multi-scale geometric features and introduces, for the first time, a truncated relative reliability loss function that explicitly models point-wise uncertainty to suppress the influence of unreliable predictions. Evaluated on the Oxford RobotCar and NCLT datasets, the proposed method achieves relative reductions in position error of 24.1% and 73.9%, respectively, significantly outperforming state-of-the-art approaches.

Technology Category

• Search and Optimization: Learning to Search
• Intelligent Robots: Localization, Mapping, and Navigation
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

LiDAR relocalization has attracted increasing attention as it can deliver accurate 6-DoF pose estimation in complex 3D environments. Recent learning-based regression methods offer efficient solutions by directly predicting global poses without the need for explicit map storage. However, these methods often struggle in challenging scenes due to their equal treatment of all predicted points, which is vulnerable to noise and outliers. In this paper, we propose LEADER, a robust LiDAR-based relocalization framework enhanced by a simple, yet effective geometric encoder. Specifically, a Robust Projection-based Geometric Encoder architecture which captures multi-scale geometric features is first presented to enhance descriptiveness in geometric representation. A Truncated Relative Reliability loss is then formulated to model point-wise ambiguity and mitigate the influence of unreliable predictions. Extensive experiments on the Oxford RobotCar and NCLT datasets demonstrate that LEADER outperforms state-of-the-art methods, achieving 24.1% and 73.9% relative reductions in position error over existing techniques, respectively. The source code is released on https://github.com/JiansW/LEADER.

Problem

• LiDAR relocalization
• 6-DoF pose estimation
• noise and outliers
• point-wise reliability
• geometric representation

Innovation

• LiDAR relocalization
• geometric encoder
• multi-scale features
• reliability-aware loss
• 6-DoF pose estimation