LighterBEV: LiDAR Global Localization Meets Online Learning
BinHong Liu, Tao Yang, Haoji Cao, Shuqi FU, YangWang Fang, Zhi Yan
AI summary
Problem
Existing deep-learning LiDAR localization methods are computationally heavy and degrade significantly when deployed in new environments due to domain shifts.
Approach
The method compresses BEV-based LiDAR features using a PCA-initialized projection module and integrates a fixed-size online buffer to continuously update model parameters from streaming data.
Key results
- Achieves state-of-the-art place recognition accuracy on KITTI and NCLT benchmarks
- Reduces feature dimensionality by fourfold while preserving rotation equivariance
- Enables rapid post-deployment adaptation to domain shifts without offline retraining
- Runs in real-time on resource-constrained hardware with low GPU memory usage
Why it matters
It allows autonomous robots to maintain accurate localization in dynamic, real-world settings without costly offline retraining or heavy computational overhead.
Abstract
LiDAR-based global localization provides accurate robot pose estimates against a prior map. Existing deep-learning methods, however, demand heavy computation and long training or inference times and degrade sharply when faced with domain shifts. This letter presents LighterBEV, a lightweight, fast, and generalizable localization method. An Informative Compression Module achieves a fourfold reduction in local-feature dimen- sionality while improving accuracy. We further integrate online learning to enable rapid post-deployment adaptation, mitigating degradation under distribution shift. Extensive experiments on four large-scale datasets show that LighterBEV achieves state-of-the-art performance with limited training data, maintains high accuracy under domain shift, and runs in real time on resource-constrained hardware—supporting both inference and online updates. To our knowledge, LighterBEV is the first LiDAR global localization approach to incorporate online learning for automatic adaptation to new environments, thereby narrowing the domain gap. Code will be released at: https://github.com/npu- ius-lab/LighterBEV.