Prioritized Multi-Agent Path Finding for AVPC Leveraging Distributional Dueling DQN

The proliferation of vehicle ownership in major cities and urban areas has presented significant challenges in parking environments, including a demand-supply imbalance of parking spaces, traffic congestion, and overall parking inconvenience. Automated Valet Parking (AVP) has emerged as a promising solution to address the parking problem by efficiently automating parking operations and optimizing parking space utilization. However, most existing AVP solutions overlook the growing popularity of Electric Vehicles (EVs) and the added complexity of EV-charging, which exacerbates the existing parking problem. Moreover, multiple vehicles attempting to find their allocated parking spaces and/or charging units in a constrained parking lot are susceptible to collisions and path conflicts. This paper proposes a novel prioritized multi-vehicle path-finding framework for Automated Valet Parking and Charging (AVPC) that leverages Distributional Dueling Deep Q-Network (TDQN). Specifically, we formulate the multi-vehicle path-finding problem as a Priority-based Distributed Cooperative Control (PDCC) optimization by assigning parking resources (parking spaces and charging units) with different priorities to multiple vehicles, each acting as an agent. We design a new reward mechanism to ensure efficient task execution among cooperating vehicles, thereby improving the parking success rate and mitigating congestion. Simulation results and analysis indicate that the proposed framework not only enhances vehicle cooperation but also significantly reduces parking space requirements, providing new ideas for smarter urban parking solutions.