A Self-Evolving Multi-Agent Robotic Swarm with Reinforcement Learning and Quantum Firefly-Ant Colony Optimization for Adaptive Environmental Monitoring

This paper presents a self-evolving multi-agent robotic swarm for adaptive environmental monitoring in unstructured and dynamic terrains, addressing limitations of existing swarms that rely on static policies and centralized control. Integrate Vision Transformers (ViTs) within a decentralized multi-agent reinforcement learning (MARL) framework, enabling each robot to extract robust spatial and temporal features from on-board visual data, essential for complex task adaptation To optimize swarm behaviour and task allocation dynamically, develop a Hybrid Quantum Firefly-Ant Colony Optimization (HQFF-ACO) algorithm that merges firefly-inspired attraction and ant pheromone-based exploration enhanced by quantum-inspired encoding for faster convergence and better avoidance of local optima. The primary objective is to achieve scalable autonomy, efficient terrain coverage, and resilient obstacle avoidance. Extensive simulations and real-world field trials show that our swarm achieves over 20 % increase in coverage efficiency and 30 % faster task completion compared to baseline MARL-controlled and heuristic-optimized swarms. These gains stem from the ViT-driven feature extraction that improves decision-making in visual-rich environments, and the HQFF-ACO module that evolves optimal task assignments in real-time. The results demonstrate that this hybrid learning and optimization approach fosters robust, emergent cooperation without global coordination making the system well-suited for environmental monitoring applications such as habitat mapping, pollution detection, and disaster assessment. Proposed system paves the way for scalable, intelligent robotic swarms capable of real-time adaptation in complex, real-world environments.