IMPLEMENTATION OF MACHINE LEARNING METHOD ON AUTONOMOUS CAR

• Main Author: Tsabitha (13314016), Shadiq Hassan Heyder (13314046) , Farah
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/27156
• Institution: Institut Teknologi Bandung

<p align="justify">The machine learning method is a method that allows the machine to learn from what has been done and aims so that the machine can make decisions to achieve better results. In this study, an autonomous car was used as a moving machine on the path that had been made. In order for the autonomous car to move following the existing trajectory, it takes a visual processing system and control program as a command to move automatically. <br /> <br /> <br /> There are two stages in designing autonomous car controls. First, a visual reading system, and second, a machine learning system. The visual reading system is done using a camera module from Raspbery pi, where the resulting image is a black and white image that has a resolution of 320 x 240 pixels. Image is processed with gaussian blur and canny edge detection to only display object boundaries. <br /> <br /> <br /> The machine learning system trains the autonomous car on the track by doing neural network training to determine the optimal weight of the artificial neural network algorithm. Each pixel of the camera image is used as a node from the first layer in the artificial neural network. In the end, the resulting outputs are three commands, front, left, and right. <br /> <br /> <br /> The implementation of the machine learning artificial neural network method in the autonomous car is obtained best after the 20th epoch with an accuracy rate of 0.8605 and a loss of 0.4273. This result is obtained in neural network training with image processing resolution of 320x120 pixels with edge detection. Autonomous car can follow the path created with average travel time of 11.88 seconds and maximum travel time of 52 seconds before making a mistake. <p align="justify"> <br />