Control of Autonomous Vehicles Based on PPO Algorithm

This research is committed to the control of driverless vehicles based on PPO algorithm, and aims to build an efficient and stable auto drive system by integrating efficient environmental awareness technology and advanced decision-making planning methods. In the study, cameras were used and the YOLOv8n algorithm was used for object detection and localization. The real-time data obtained was used to train the vehicle's decision model, enabling it to develop a reasonable driving path based on the current obstacle and target point positions, and select the appropriate direction of travel accordingly. This study implemented strict data preprocessing, including the application of data augmentation techniques, to ensure the quality of modeling data and the generalization ability of the model. In the model training and validation phase, we built an embedded car control simulation platform, controlled the car through Raspberry Pi 4B, and imported the pre trained model. After 2000 rounds of training, the model achieved stable convergence and was able to correctly avoid obstacles and reach the target point, basically achieving the expected goal. The experimental results show that the PPO algorithm has fast convergence speed and strong robustness while maintaining learning stability. This study not only provides theoretical basis and technical support for the development of autonomous driving technology, but also contributes to promoting social awareness and acceptance of autonomous driving technology, and provides solutions for the safe popularization and widespread application of autonomous vehicles.