OCLPlace: Online Continual Learning on LiDAR Streams for Place Recognition
BinHong Liu, Kaixiao Ye, YangWang Fang, Zhi Yan, Tao Yang
AI summary
Problem
Existing deep learning-based LiDAR place recognition methods degrade significantly when deployed in unseen or rapidly changing environments, and offline fine-tuning is too slow and prone to catastrophic forgetting.
Approach
OCLPlace introduces an online continual learning framework with a dual-memory system: a short-term memory for fast consolidation and hardness-aware forgetting, and a long-term memory for durable replay of hard triplets, enabling on-the-fly adaptation to streaming LiDAR data.
Key results
- First online continual learning framework for LiDAR place recognition
- Dual-memory architecture balances rapid adaptation with forgetting resistance
- Validated across six large-scale ground, aerial, and cross-sensor datasets
- Publicly released source code and standardized benchmarks
Why it matters
It enables autonomous robots to maintain reliable long-term localization in dynamic real-world environments without costly offline retraining or manual annotations.
Abstract
LiDAR place recognition is a critical component of LiDAR-based localization pipelines, tasked with identify- ing previously visited places across diverse environments and temporal conditions. A growing body of deep learning–based approaches has recently tackled this problem. However, their performance often degrades when the models are deployed in unseen environments. Although offline fine-tuning can partly recover performance, it is prone to catastrophic forgetting of previously acquired knowledge and cannot respond quickly enough to rapidly changing data distributions. In this paper, we introduce OCLPlace, an online continual learning frame- work that learns directly from highly temporally correlated LiDAR streams and strikes a trade-off between rapid domain adaptation and resistance to catastrophic forgetting. To the best of our knowledge, OCLPlace is the first LiDAR place- recognition approach enhanced by online continual learning that can automatically adapt to new environments while mit- igating catastrophic forgetting. Experimental results on six large-scale datasets, which cover both ground-view and aerial- view scenarios, demonstrate the effectiveness and robustness of our method. The source code will be publicly available at: https://github.com/npu-ius-lab/OCLPlace.