Gaussian Mixture Model and Deep Reinforcement Learning Based Driving Robot System Gearshift Strategy With Vehicle Longitudinal Driving States

摘要

Vehicle tests play a very important role in automotive manufacturing, and by using a driving robot for vehicle tests, the efficiency and accuracy of vehicle tests can be improved. In order to improve the accuracy and adaptability of the gearshift strategy for driving robot system (DRS) in different manipulated vehicle longitudinal driving states, a gearshift strategy of DRS based on the Gaussian mixture model (GMM) and deep reinforcement learning (DRL) is proposed. First, a dynamics model of DRS is established. Then, a GMM-based vehicle driving state identification model of DRS is proposed. The general features of vehicle driving data are extracted by autoencoder algorithm. The clustering of vehicle longitudinal driving states is made by the GMM. Finally, DRL-based gearshift schedules of DRS are proposed. A reinforcement learning model of gearshift schedule for DRS is established according to the Markov decision process (MDP). The state space, action space, and reward rule of MDP are designed. The gearshift schedules of DRS in different longitudinal driving states are solved. The proposed gearshift strategy of DRS is verified through simulation and testing in different driving cycle tests. Simulation and testing results show the effectiveness of the proposed strategy.