B-DART: A Beagle-Inspired Dynamic Agent Response Framework for Enhanced Disaster Management

Abstract

Static, inflexible dispatch protocols are critically failing the dynamic and concurrent demands put on emergency response systems by large-scale disasters. This work introduces B-DART, a novel multi-agent optimization framework that seeks to overcome these limitations through biologically inspired, experience-based learning. B-DART captures the memory, exploration, and exploitation behaviors of hunting beagles into a cohesive algorithm, BIOA, for dynamic team formation. We extensively test this framework within a simulated 20x20 grid environment modeling the operational space of San Francisco using a heterogeneous fleet of 17 agents managing stochastically generated high-priority emergencies. Simulation results at 50 time steps demonstrate that the proposed framework can indeed achieve 100% agent utilization and learn through the addition of 3 belief entries. However, even more surprising are the huge operational deficiencies that come to light: out of the 15 spawned tasks, only 4 get completed for a success rate of 26.67% at an average response time of 15.5 steps; medical emergencies are not attended to at all. While B-DART is successful in validating a novel architecture for adaptive coordination, current performance highlights the critical need for enhanced priority handling and resource scalability mechanisms in complex disaster scenarios.