DESIGN AND IMPLEMENTATION OF ROAD FOLLOWING ON A SELF-DRIVING CAR PROTOTYPE BASED ON JETSON NANO WITH 5G* NETWORK

• Main Author: Maitsa Gunawan, Syaghina
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/82259
• Institution: Institut Teknologi Bandung

The development of self-driving car technology offers great potential in enhancing transportation efficiency and safety. Self-driving cars can reduce human involvement in driving, potentially lowering traffic accident rates often caused by human error. This technology is also expected to optimize travel and reduce congestion by utilizing more efficient navigation systems. Currently, the self- driving car industry is experiencing rapid growth worldwide, including in Indonesia, where the implementation of this technology may face various infrastructure and regulatory challenges. This research aims to develop a self-driving car prototype based on Jetson Nano by utilizing Convolutional Neural Network (CNN) algorithms for steering angle prediction. In this study, the CNN ResNet18 architecture was used for the steering angle prediction model, which was then trained and tested with a track image dataset. The system combines the JetRacer AI Kit and the IMX219 camera module for real-time visual data acquisition and processing. The model was trained with a dataset that includes simple track conditions and tested on two simulated tracks, simple and complex ones. This approach aims to enable the system to follow the track by providing accurate and responsive predictions to various road conditions, thus enhancing driving safety and comfort. The research results indicate that the CNN ResNet18 model successfully underwent the training process. This is evidenced by the stable and closely aligned training and validation loss graphs. In the road-following simulation tests, the model demonstrated stability with smooth and non-erratic steering angle adjustments. On track (a), JetRacer exhibited robust performance in following the path, whether moving clockwise or counterclockwise. On track (b), the model continued to follow the path effectively, showcasing good generalization capabilities under varying track conditions. Overall, the model demonstrated stable performance and is ready for deployment on various tracks with minimal further adjustments. This research concludes that the self-driving car technology based on Jetson Nano and CNN can be effectively implemented using WiFi. Although the integration of 5G technology has not been successful due to its limited availability, the system demonstrated acceptable responsiveness for the road following feature in the self- iv driving car. According to the results obtained, the ResNet18 CNN model is capable of accurately predicting the steering angle, though there are still some challenges on more complex tracks. Despite not matching the speed and stability expected from 5G technology, the WiFi connection performed sufficiently well to support the basic operations of JetRacer in following the track. For further development, it is recommended to expand the dataset with more varied and complex tracks to improve the model's generalization capabilities. Additionally, the integration of 5G technology to enhance system responsiveness and reliability should be considered, along with the addition of sensors such as LiDAR or stereo cameras to improve object detection and environmental information. The implementation and results of this research are expected to make a significant contribution to the development of self-driving car technology and serve as a foundation for future research in the same field. It is hoped that the research results can be applied and further developed to support technological advancements in the future, especially in the field of self-driving cars in Indonesia. Thus, this technology is expected to help address various transportation problems, increasing efficiency, safety, and user comfort in the future. Overall, this research demonstrates the great potential of using deep learning in developing sophisticated and reliable self-driving car systems.