STUDY AND IMPLEMENTATION OF AUTONOMOUS VEHICLE LATERAL CONTROL SYSTEM VIA LYAPUNOV STABILITY METHOD AND DEEP REINFORCEMENT LEARNING

• Main Author: Apeco Putra, Dimas
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/55241
• Institution: Institut Teknologi Bandung

Autonomous vehicle (AV) is a solution to reduce traffic jams and accident rates due to driver error. AV is a car that can navigate without human intervention. One of the essential subsystems of an AV is lateral control system, which functions to direct the car to the desired path through manipulation of the car's steering angle. Stability, nonlinearity, and optimal performance are common challenges in designing control systems, including AV lateral control systems. Lyapunov stability method is one of the methods for designing a nonlinear system controller. Deep reinforcement learning (DRL) is a method that is widely used and developed in this decade to find optimal policy solutions in the case of dynamic programming. In this study, a study on the use of DRL was carried out to optimize the performance of the Lyapunov stability method controller in the lateral control system for AV. Stanley lateral controller, which is popular in the AV literature, used as the performance baseline. Controller performance is tested through simulation and real-scale implementation. The simulation test uses the CARLA simulator and the kinematic simulator designed in this research. From the test, it was found that the Lyapunov-DRL controller consistently produced better performance than the Stanley controller, with decrease in IAE, ISE, ITAE, and RMSE values of 80.72%, 57.32%, 92.86%, and 34.64%, respectively. Real-scale implementation test was done using a modified Yamaha YDRE 2011 golf cart. From the test, it was found that the Lyapunov-DRL controller produced better performance than the Stanley controller with decrease in IAE, ISE, ITAE, and RMSE values of 37.22%, 43.34%, 43.72%, and 24.43%, respectively.