Emulating Emergency Vehicle Navigation for Mixed Traffic Environment Using Graph Prediction and Simulation

Abstract

Emergency vehicles (EVs) must move through areas with mixed traffic as fast as they can in order to deliver aid in a timely manner. The dynamic and unpredictable nature of traffic makes this a difficult task. In some bottleneck situations, such as highway ramping and intersections, reinforcement learning (RL) has been used to simulate traffic and improve EV navigation. However, the complexity of real-world traffic environments is not fully captured by RL-based simulations, which are frequently restricted to a small number of entities. Our paper proposes a novel method to simulate EV navigation in mixed-traffic environments using SUMO simulation that utilizes a graph neural network (GNN) algorithm called Graph SAGE to learn node representations or the representations of the relationships between nodes in a graph through training, and a custom action policy function that uses the link states that we predict using Graph SAGE to determine the EVs' next course of action and makes decisions about lane changes, speed adjustments, and acceleration based on the predicted link states, vehicle types, and relative positions of vehicles. We demonstrate that, when compared to baseline approaches, our approach can significantly improve the navigation efficiency of EVs by achieving better speeds and reducing overall and emergency waiting times. Our findings demonstrate how well graphs model and simulate intricate traffic situations.