Deep Learning Based Multi-Zone AVP System Utilizing V2I Communications

Abstract

Autonomous Valet Parking (AVP) is a technology that enables vehicles to park themselves without human intervention. It uses advanced sensing and communication systems to find a suitable parking space and to park the vehicle safely and efficiently. While various artificial intelligence (AI) based methods have demonstrated the benefits of AVP, including reducing traffic congestion, improving safety, and enhancing convenience and comfort for drivers, the issue of developing and evaluating AVP systems that can effectively handle multi-zone parking areas in real-world settings is yet to be solved. This paper presents an AVP system for three parking zones situated within a 1 km radius and utilizes a combination of existing tools and Deep Deterministic Policy Gradient (DDPG) algorithm to address the issue. DDPG algorithm controls the AVP system to allocate parking spaces efficiently in order to navigate and park vehicles autonomously. This work assumes the utilization of 5G-NR Vehicle-to-Infrastructure (V2I) communications for information exchange between vehicles and the system. It also studies the effect of communication latency on the system performance. Results of simulations show that the proposed system efficiently and safely parks vehicles in the three parking zones, achieving a reduction of 7% in waiting time compared to existing deep reinforcement learning methods. This work represents a notable advancement over current solutions and helps to advance the vision of smart cities for the future.

Keywords

• Autonomous valet parking
• multi-zone AVP
• deep deterministic policy gradient
• vehicle-to-infrastructure communications