Deep Reinforcement Learning and Fuzzy Logic-Based Adaptive Shift Control Method of Driving Robot System

In order to improve the fuel economy of the test vehicle manipulated by the driving robot, an adaptive shift control method based on deep reinforcement learning (DRL) and fuzzy logic of driving robot system is proposed. The driving robot system (DRS) is installed in the cockpit of a vehicle, which does not require any modifications to the vehicle. It includes a physical robot that mechanically operates vehicle and a controller for an automated driving. Firstly, the dynamic model of driving robot system is established. Then, a deep reinforcement learning model of gearshift strategy is established. Aiming at fuel economy, the throttle opening, vehicle speed and vehicle acceleration are used as the observation parameters of DRL model for gearshift strategy. Then, a nonlinear disturbance observer is established to observe the parameters of the DRL model. Furthermore, an adaptive shift controller consisting of gearshift strategy and fuzzy logic is proposed considering the Gaussian pulse secondary impact load model during the shifting process. The continuous control of DRS is realized by the fuzzy logic controller. Finally, the stability proof of the proposed shift control method is conducted. The proposed shift control method is verified through testing in different driving cycle tests. Test results demonstrate the effectiveness of the proposed shift control method.