A Neuromorphic Spiking Neural Network-Based Continual Learning Framework for Energy-Efficient Edge Intelligence in Autonomous UAV Swarm Navigation

Abstract

The rapid advancement of autonomous Unmanned Aerial Vehicles (UAVs) has significantly transformed numerous civilian, industrial, and defense applications including precision agriculture, environmental monitoring, disaster management, surveillance, search and rescue operations, intelligent transportation, smart city monitoring, and military reconnaissance. Modern UAV swarm systems operate collaboratively to accomplish complex missions through distributed sensing, cooperative decision-making, adaptive path planning, and decentralized communication. Unlike single-drone systems, autonomous UAV swarms offer enhanced robustness, scalability, fault tolerance, and mission efficiency by enabling multiple UAVs to coordinate dynamically in uncertain environments. However, realizing real-time intelligence on resourceconstrained edge devices remains a significant challenge because conventional deep learning models demand substantial computational resources, memory capacity, and energy consumption, which exceed the limitations of lightweight onboard embedded processors. Furthermore, continuously changing operational environments require UAVs to learn new navigation behaviors without forgetting previously acquired knowledge, making continual learning an essential capability for long-term autonomous swarm intelligence. This paper proposes a Neuromorphic Spiking Neural Network-Based Continual Learning Framework for energy-efficient edge intelligence in autonomous UAV swarm navigation. The proposed framework integrates event-driven neuromorphic sensing, Spiking Neural Networks, Continual Learning, Graph-based Swarm Coordination, Reinforcement Learning, Neuromorphic Edge Computing, Explainable Artificial Intelligence, and adaptive knowledge consolidation into a unified autonomous navigation architecture. Event-based visual sensors continuously capture environmental changes and convert them into asynchronous spike streams processed by energy-efficient neuromorphic hardware. The continual learning module dynamically adapts navigation policies while preserving previously acquired swarm behaviors through synaptic consolidation and memory replay mechanisms. Experimental evaluation demonstrates that the proposed framework significantly improves navigation accuracy, continual learning capability, energy efficiency, obstacle avoidance performance, swarm coordination, computational scalability, and mission robustness compared with conventional deep learning-based UAV navigation systems. The proposed architecture establishes a scalable and intelligent foundation for future autonomous edge computing in collaborative UAV swarms.