Unlocking Generalization Power in LiDAR Point Cloud Registration

2025-03-13

AI Summary

Existing LiDAR point cloud registration methods exhibit insufficient generalization across varying distances and disparate datasets, compromising safety-critical applications such as autonomous driving. To address this, we propose a lightweight, generalization-oriented Transformer architecture: (i) we eliminate cross-frame attention to enhance out-of-distribution robustness; (ii) we introduce a progressive self-attention module to mitigate structural ambiguities in large-scale scenes; and (iii) we fuse bird’s-eye-view (BEV) semantic features to improve geometric–semantic consistency. Trained in a fully unsupervised manner, our method achieves state-of-the-art generalization performance: 94.5% and 91.4% registration recall on KITTI and nuScenes for cross-distance evaluation, respectively, and 90.9% recall for cross-dataset transfer from nuScenes to KITTI—surpassing all prior methods.

Technology Category

• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

In real-world environments, a LiDAR point cloud registration method with robust generalization capabilities (across varying distances and datasets) is crucial for ensuring safety in autonomous driving and other LiDAR-based applications. However, current methods fall short in achieving this level of generalization. To address these limitations, we propose UGP, a pruned framework designed to enhance generalization power for LiDAR point cloud registration. The core insight in UGP is the elimination of cross-attention mechanisms to improve generalization, allowing the network to concentrate on intra-frame feature extraction. Additionally, we introduce a progressive self-attention module to reduce ambiguity in large-scale scenes and integrate Bird's Eye View (BEV) features to incorporate semantic information about scene elements. Together, these enhancements significantly boost the network's generalization performance. We validated our approach through various generalization experiments in multiple outdoor scenes. In cross-distance generalization experiments on KITTI and nuScenes, UGP achieved state-of-the-art mean Registration Recall rates of 94.5% and 91.4%, respectively. In cross-dataset generalization from nuScenes to KITTI, UGP achieved a state-of-the-art mean Registration Recall of 90.9%. Code will be available at https://github.com/peakpang/UGP.

Problem

• Enhancing generalization in LiDAR point cloud registration.
• Addressing limitations in cross-distance and cross-dataset performance.
• Improving safety in autonomous driving via robust registration methods.

Innovation

• Eliminates cross-attention for better generalization
• Introduces progressive self-attention for scene clarity
• Integrates BEV features for semantic scene understanding