Eco-Friendly Disaster Evacuation Framework for 6G Connected and Autonomous Vehicular Networks

Abstract

Crowd and evacuation management have been active areas of research and study over recent years. Connected Autonomous Vehicles (CAVs) could provide safe and vital life-saving reinforcement when evacuating populations from a large-scale disaster site. However, navigation in disaster-affected environments is vulnerable to the presence of unpredictable moving obstacles. Existing solutions perform optimal route determination while considering the presence of a single rescue vehicle. To optimize path planning and enhance energy efficiency in the presence of a fleet of rescue vehicles and Unmanned Aerial Vehicles (UAVs), multiple sources and target destinations, we propose a Multi-Agent Deep Reinforcement Dijkstra (MADRD) algorithm in a sixth generation (6G) assisted environment to achieve cooperative navigation in the Route Planning Framework. The MADRD algorithm allows vehicles to learn from the experiences and behavior of other CAVs to obtain an optimal path from the source to the affected target site. Route selection and throttle parameters are optimized to achieve efficient fuel consumption, reduced travel time and lower traffic congestion. We simulate the Route Planning Framework on a large-scale real-world road network. The experimental results indicate that MADRD achieves 40% less fuel consumption, 45% higher road throughput and 34% less evacuation time than its state-of-the-art counterparts.