Towards Lifelong Aerial Autonomy: Geometric Memory Management for Continual Visual Place Recognition in Dynamic Environments
2026-04-10 • Robotics
RoboticsComputer Vision and Pattern RecognitionMachine Learning
AI summaryⓘ
The authors focus on improving how drones remember places when flying over changing environments for a long time. They treat the problem as learning new information from different missions without forgetting old knowledge, which is hard because geographic features can look very different. To solve this, they created a system that separates permanent map info from recent experiences and smartly picks what to remember based on how varied the data is. Their tests show this helps drones learn better and keep important knowledge without getting confused by new data. They found that keeping a diverse set of features is more helpful than focusing on how hard samples are, especially in complex, changing outdoor areas.
Visual Place RecognitionContinual LearningDomain-Incremental LearningCatastrophic ForgettingAerial AutonomyMemory BufferSpatial GeneralizationExperience ReplayStructural DiversityGeographic Features
Authors
Xingyu Shao, Zhiqiang Yan, Liangzheng Sun, Mengfan He, Chao Chen, Jinhui Zhang, Chunyu Li, Ziyang Meng
Abstract
Robust geo-localization in changing environmental conditions is critical for long-term aerial autonomy. While visual place recognition (VPR) models perform well when airborne views match the training domain, adapting them to shifting distributions during sequential missions triggers catastrophic forgetting. Existing continual learning (CL) methods often fail here because geographic features exhibit severe intra-class variations. In this work, we formulate aerial VPR as a mission-based domain-incremental learning (DIL) problem and propose a novel heterogeneous memory framework. To respect strict onboard storage constraints, our "Learn-and-Dispose" pipeline decouples geographic knowledge into static satellite anchors (preserving global geometric priors) and a dynamic experience replay buffer (retaining domain-specific features). We introduce a spatially-constrained allocation strategy that optimizes buffer selection based on sample difficulty or feature space diversity. To facilitate systematic assessment, we provide three evaluation criteria and a comprehensive benchmark derived from 21 diverse mission sequences. Extensive experiments demonstrate that our architecture significantly boosts spatial generalization; our diversity-driven buffer selection outperforms the random baseline by 7.8% in knowledge retention. Unlike class-mean preservation methods that fail in unstructured environments, maximizing structural diversity achieves a superior plasticity-stability balance and ensures order-agnostic robustness across randomized sequences. These results prove that maintaining structural feature coverage is more critical than sample difficulty for resolving catastrophic forgetting in lifelong aerial autonomy.