A Grid-Based Reinforcement Learning Approach forAmbulance Dispatching

Abstract

Ambulance dispatching for critical life risk emergencies (type-1) poses significant challenges in rural and resource-constrained settings, where ensuring both rapid response and long-term coverage is critical. Existing methods often overlook ambulance type constraints and focus narrowly on minimizing immediate response time. In this paper, we propose a dispatching strategy called GARD (Grid-aware Reinforcement Dispatching), which partitions the service area into grids categorized by historical emergency intensity. A reinforcement learning agent uses this structure to select ambulances, balancing proximity with anticipated system-wide impact. Experiments based on real-world EMS data demonstrate that GARD improves performance metrics for high-priority emergencies while preserving better future coverage.