SIMULATION OF LATERAL CONTROLLER USING CAMERA FOR AUTONOMOUS CAR LANE FOLLOWING SYSTEM

• Main Author: Putu Gede Amartya K.S., Ngakan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/74565
• Institution: Institut Teknologi Bandung

Autonomous cars, also known as autonomous vehicles, have been a topic that has gained increasing attention in recent years. Human negligence is the leading cause of traffic accidents, whether due to fatigue, lack of focus, or other human factors. Autonomous cars have the ability to reduce this risk because they use technology such as sensors, computers, and software that are able to respond better and are not affected by emotional distractions or fatigue. One important aspect of driving that autonomous cars must do well is lane following. Generally, driving lanes are defined by the presence of road markings that serve as visual cues. One of the advantages of road markings is that they have been in use for a long time and are commonly used in many places, so autonomous cars will be easier to implement if autonomous cars can follow lanes with existing infrastructure. In this final project, a study is conducted that simulates a lane following system using an RGB camera and a lateral controller to keep the car in the lane. Simulation is already a common way to design and test an autonomous car system, one of its advantages is to speed up the testing of an idea. In this research, the simulation is done in Carla Simulator software and in conjunction with ROS (Robot Operating System). Carla Simulator serves to simulate physical phenomena while ROS serves for communication between processes in the path follower system. The lane following system is designed by processing the image of the car's front-facing camera to detect the presence of road markings that demarcate the autonomous car's lane. From this information, the position of the car relative to the road can be determined, the representations used in this research are lateral error and orientation error, which respectively express the closest distance of the car to the lane, and the angle formed from the difference between the lane orientation and the car orientation. The error information will be used for the lateral controller to determine the steering angle that can reduce the detected error. This whole process forms a feedback process or closed loop that will keep the moving car on the track. The lateral controller used in this research is the Stanley Controller, a non-linear controller that has been shown to have global stability. Generally, Stanley controllers use various combinations of sensors to provide the error information needed to determine the steering angle. However, since each sensor is not always reliable, the Stanley controller needs to be tested when it can only use one sensor, and in this final project research, a camera is chosen as the sensor, because road markings are found in various places as a visual lane funding, as well as easy access to RGB cameras. Related research in the Engineering Physics Study Program, has also designed an autonomous car with a lateral stanley controller, but the camera as a sensor is a novelty carried out in this final project research. The camera-based error detection subsystem was tested before operating to provide errors to the lateral controller. Error Detection is divided into tests to detect lateral errors and orientation errors, and for lateral error testing in the range of measurement values obtained a sensitivity value of 0.67, a maximum measurement of 1 meter, a minimum measurement of -1 meter, an average ME (Mean Error) of - 0.065 with a standard deviation of 0.208 and an average MAE (Mean Absolute Error) of 0.183 with a standard deviation of 0.127. Then testing the orientation error detector in the measurement value range gives a sensitivity value of 0.68, a maximum measurement of 8?, a minimum measurement of -8? meters, an average ME of -0.675 with a standard deviation of 2.225 and an average MAE of 3.186 with a standard deviation of 1.395. The error detector test concluded that there was bias and slope drift from the actual measurement value. So that calibration is carried out on the basis of the sensitivity and ME values for each error detector, so that for the lateral error detector, the value after calibration is obtained, the sensitivity is 1.03, the average ME is 0.043 with a standard deviation of 0.068, and the average MAE is 0.264 with a standard deviation of 0.126. Then for the orientation error detector after calibration, the sensitivity is 1.12, the average ME is 0.142 with a standard deviation of 0.047 and the average MAE is 4.546 with a standard deviation of 2.089. The lane tracking system was tested on 2 different trajectory shapes, the first trajectory is a straight path followed by a left turn and ending with a straight path. While the second trajectory is a constant right turn road. The Path Follower System has been tested and successfully followed the path on the first and second trajectories. And provides RMSE and MAE values for lateral errors of [0.139; 0.145] meters for the first trajectory, [0.106; 0.132] meters for the second trajectory and orientation errors of [0.974; 0.797]? for the first trajectory, and [0.610; 0.558]? for the second trajectory. The lane following system was also tested for initial driving errors. On each track the lane following system was tested for maximum positive and negative lateral initial errors, and maximum positive and negative orientation initial errors. From all these tests, the lane following system successfully corrected the initial driving errors, and stayed within the lane during driving. On the first track, the RMSE and MAE of lateral error and orientation error were [0.230; 0.167] meters, [1.911; 1.271] ?. On the second track, the RMSE and MAE of lateral and orientation errors are [0.169; 0.153] meters, [1.447; 0.956] ?.