A Reinforcement Learning based Path Guidance Scheme for Long-range Autonomous Valet Parking in Smart Cities

Abstract

Finding a parking slot in the city centre has always been a great challenge. In many cases, drivers spend a lot of time roaming around looking for an empty and suitable parking slot. The emerging machine learning technologies in intelligent transport system has made it more flexible for Electric Autonomous Vehicle (EAV) to find a parking slot and get parked. The Long-range Autonomous Valet Parking (LAVP) allows an EAV to drop user at a suitable drop-off spot and select an economical parking slot. With the evolution of battery operated vehicles, the primary concern is efficient use of battery resources. This can be done either by maximizing battery capacity or by smartly using battery with existing capacity. During the parking process, most of the energy is consumed by finding an optimal path to parking slot. The work proposed in this paper guides EAV from a random starting point to nearest drop-off spot and CP. A Reinforcement Learning based Autonomous Valet Parking technique (RL-LAVP) has been designed to guide EAV to drop-off spot, CP and minimize the total distance covered during this process. The RL-LAVP results show a significant improvement towards minimizing covered distance and consumed energy when compared with RaNdom (RN) parking and LAVP parking techniques.