A Hybrid Multi-Agent Emergency Response Simulation Platform with Centralized Dispatch and Localized Agent Autonomy

Abstract

Urban emergency response systems are complex, high-risk, and hard to evaluate directly in real-world settings due to operational limits, costs, and safety concerns. Just as circuit designers use virtual prototyping platforms like Tinkercad before deploying physical hardware, this work introduces a simulation-first approach for planning and evaluating emergency responses. We develop a hybrid multi-agent simulation platform that allows for "test-before-deploy" experimentation for ambulance dispatch and urban incident management policies. The proposed system is built on a realistic urban road network derived from OpenStreetMap and runs using SUMO (Simulation of Urban Mobility) with TraCI integration. It models civilian traffic, hospital infrastructure, random incident generation, and a fleet of ambulances acting as autonomous agents. Each ambulance keeps localized state, intent, and behavior policies, allowing for decentralized decision-making for patrol and response. A centralized dispatch layer coordinates incident assignments using a scoring system, which can be enhanced with a tabular reinforcement learning policy for selecting primary responders.

The platform adds more operational realism through severity-based responder allocation, scene dwell times, hospital handover delays, hotspot-aware patrol rebalancing, and localized traffic-signal preemption to simulate green corridors. An expandable experimental layer allows for synthetic modeling of weather conditions, traffic congestion patterns over time, variations in district-level demand, and hospital operational limits. The system provides structured operational metrics, including response times, completion rates, and resource use, which are analyzed through a dashboard-driven process. This setup allows for systematic comparison of dispatch strategies, fleet configurations, and environmental conditions under controlled, reproducible scenarios.

The main contribution of this work is a modular and rollback-safe simulation platform that connects heuristic multi-agent control with learning-based dispatch in a realistic urban environment. By allowing rapid prototyping and evaluation of emergency response strategies before real-world deployment, the platform acts as a digital testbed for research in smart-city mobility, multi-agent coordination, and optimizing emergency logistics.