STUDY OF LATERAL INTEGRAL BACKSTEPPING CONTROL WITH UNSCENTED KALMAN FILTER AS LOCALIZATION SYSTEM FOR AUTONOMOUS CAR

• Main Author: Vardani Rulianto, Muhammad
• Format: Final Project
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/60986
• Institution: Institut Teknologi Bandung

The lateral control system and the localization system are important parts of an autonomous car. Previous research has explored these two things separately. There has never been an analysis related to the localization system and the lateral control system simultaneously. Yet the two systems mutually influence each other. This study aims to analyze the influence of the localization system on the performance of lateral control by designing and testing. The method used is to design a localization system based on the Unscented Kalman Filter (UKF). As a variation of the controller, the Integral Backstepping controller is designed. The test is done by comparing the control scheme with and without UKF. For comparison, the Stanley controller is also used. The integral backstepping controller has two gain parameter, ????1 and ????2, that needs to be determined. This gain was determined by the Reinforcement Learning Soft Actor-Critic (SAC) agent who had been trained in a kinematic simulator. Likewise, UKF has ????0 and ????parameters whose values are determined by the Grey Wolf Optimizer (GWO) algorithm. The tests were carried out using hardware-in-the-loop simulation using CARLA simulator and ROS. It was found that the selection of the localization system significantly affected the performance control. The integral backstepping controller with a GNSS based localization system failed to perform the simulation. On the other hand, an integral backstepping controller with a UKF-based localization system can complete the simulation stably. The performance index at the maximum speed of 7.5 m/s for integral backstepping obtained MAE 0.268 m, RMSE 0.448 m with a simulation time of 19.870 seconds at 7.5 m/s; Stanley got an MAE of 0.652 m, RMSE of 0.943 m, and a simulation time of 23.090 seconds.