Q learning-based Autonomous Valet Parking System

Abstract

In this paper, we propose a brand new vehicle lateral motion control algorithm for autonomous parking system utilizing Q-learning algorithm. Normally for optimal vehicle control, linearization is introduced to deal with the nonlinearity of vehicle dynamics, which reduces the optimality of the derived control laws. To solve the problem, a Q-learning based vehicle parking control algorithm is proposed. A path planning method is introduced to the design of the state vector in the Q-learning algorithm for vehicle lateral control. Feasibility of the proposed algorithm is validated by computational simulation results showing satisfactory performances on the test scenario.