S-BEVLoc: BEV-based Self-supervised Framework for Large-scale LiDAR Global Localization

2025-09-10

AI Summary

To address the high annotation cost of LiDAR-based global localization—stemming from its reliance on high-precision GPS/SLAM ground-truth poses—this paper proposes the first BEV-based self-supervised end-to-end learning framework. Given only a single BEV image and its corresponding geographic coordinate, the method automatically constructs triplets by leveraging known geographic distances between keypoint regions. A SoftCos loss is introduced to enhance local feature discriminability, while a CNN-NetVLAD architecture generates robust global descriptors. Crucially, no ground-truth pose supervision is required, significantly improving scalability and deployment efficiency. Evaluated on large-scale KITTI and NCLT benchmarks, the method achieves state-of-the-art performance in both place recognition and global localization tasks, with strong loop-closure detection robustness.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Machine Learning: Unsupervised & Self-Supervised Learning
• Intelligent Robots: Localization, Mapping, and Navigation

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

LiDAR-based global localization is an essential component of simultaneous localization and mapping (SLAM), which helps loop closure and re-localization. Current approaches rely on ground-truth poses obtained from GPS or SLAM odometry to supervise network training. Despite the great success of these supervised approaches, substantial cost and effort are required for high-precision ground-truth pose acquisition. In this work, we propose S-BEVLoc, a novel self-supervised framework based on bird's-eye view (BEV) for LiDAR global localization, which eliminates the need for ground-truth poses and is highly scalable. We construct training triplets from single BEV images by leveraging the known geographic distances between keypoint-centered BEV patches. Convolutional neural network (CNN) is used to extract local features, and NetVLAD is employed to aggregate global descriptors. Moreover, we introduce SoftCos loss to enhance learning from the generated triplets. Experimental results on the large-scale KITTI and NCLT datasets show that S-BEVLoc achieves state-of-the-art performance in place recognition, loop closure, and global localization tasks, while offering scalability that would require extra effort for supervised approaches.

Problem

• Eliminates need for costly ground-truth pose supervision in LiDAR localization
• Addresses scalability limitations in supervised BEV-based localization approaches
• Solves large-scale localization without GPS/SLAM odometry dependency

Innovation

• Self-supervised BEV framework
• Geographic distance triplet training
• SoftCos loss for feature learning