Risk-Aware Multi-Agent Approach for Energy-Efficient Emergency Traffic Management

In this study, we propose a novel Risk-Aware Hierarchical Energy-Managed Twin (RA-HEMT++) framework to address these problems by leveraging coordinated UAV-UGV collaboration and deep reinforcement learning to form traffic corridors in emergency scenarios. The proposed framework combines hierarchical multi-agent reinforcement learning, spatiotemporal traffic-state fusion, digital twin simulation, and energy-aware optimization, enabling adaptive, real-time traffic management. Various datasets are used, including SUMO traffic simulations, UAV traffic observations, NGSIM vehicle trajectories, and OpenStreetMap road networks, to build a realistic and scalable urban traffic environment. The framework uses a hierarchical decision-making approach with strategic corridor planning, cooperative task allocation, and low-level motion control, while also reducing the delay time of emergency vehicles, the severity of congestion, and system energy usage. Experimental results show the superiority of the proposed RA-HEMT++ framework over traditional traffic control, RL-based optimization, federated DRL, and hierarchical RL methods. The proposed model can achieve 75 secs of corridor formation time, 55 secs of emergency vehicle delay time, a 60% reduction in congestion, and 3.0 kWh of energy consumption. The results demonstrate that RA-HEMT++ is a robust, scalable, and energy-efficient solution for next-generation emergency traffic management in smart cities.