🤖 AI Summary
To address the low energy efficiency, high latency, and poor scalability of visual place recognition (VPR) in large-scale environments, this paper proposes a Modular Spiking Neural Network (Modular SNN) architecture. The architecture introduces a novel decoupled design that enables partitioned modeling and lightweight multi-module integration—each module containing only 1,500 neurons—and is the first to systematically incorporate continuous image sequence matching. By synergistically integrating brain-inspired computation principles with neuromorphic hardware co-design, the method achieves significantly higher R@1 accuracy than conventional CNNs and SNNs across multiple VPR benchmarks. Module-level integration enhances robustness, while sequence matching yields up to a 12.3% relative performance gain. Experimental results demonstrate the proposed solution’s balanced advantages in real-time inference, ultra-low power consumption, and scalability—establishing a viable neuromorphic approach for edge robot visual localization.
📝 Abstract
In robotics, spiking neural networks (SNNs) are increasingly recognized for their largely unrealized potential energy efficiency and low latency particularly when implemented on neuromorphic hardware. This article highlights three advancements for SNNs in visual place recognition (VPR). First, we propose modular SNNs (Modular SNN), where each SNN represents a set of nonoverlapping geographically distinct places, enabling scalable networks for large environments. Second, we present ensembles of Modular SNNs, where multiple networks represent the same place, significantly enhancing accuracy compared to single-network models. Each of our Modular SNN modules is compact, comprising only 1500 neurons and 474k synapses, making them ideally suited for ensembling due to their small size. Finally, we investigate the role of sequence matching in SNN-based VPR, a technique where consecutive images are used to refine place recognition. We demonstrate competitive performance of our method on a range of datasets, including higher responsiveness to ensembling compared to conventional VPR techniques and higher R@1 improvements with sequence matching than VPR techniques with comparable baseline performance. Our contributions highlight the viability of SNNs for VPR, offering scalable and robust solutions, and paving the way for their application in various energy-sensitive robotic tasks.
