Real-World Autonomous Driving Control: An Empirical Study Using the Proximal Policy Optimization (PPO) Algorithm

Zhao Peng, Yuan Zhongxian, Thu Kyaw, 宮崎 隆彦

doi:10.5109/7183372

Real-World Autonomous Driving Control: An Empirical Study Using the Proximal Policy Optimization (PPO) Algorithm

DOI 機関リポジトリ (HANDLE) オープンアクセス

Zhao Peng

九州大学大学院総合理工学府
Yuan Zhongxian

Faculty of Environment and Life, Beijing University of Technology
Thu Kyaw

九州大学大学院総合理工学府
宮崎隆彦

九州大学大学院総合理工学府

この論文をさがす

CiNii Books

説明

This article preprocesses environmental information and use it as input for the Proximal Policy Optimization (PPO) algorithm. The algorithm is directly trained on a model vehicle in a real environment, allowing it to control the distance between the vehicle and surrounding objects. The training converges after approximately 200 episodes, demonstrating the PPO algorithm's ability to tolerate uncertainty, noise, and interference in a real training environment to some extent. Furthermore, tests of the trained model in different scenarios reveal that even when the input information is processed and does not provide a comprehensive view of the environment, the PPO algorithm can still effectively achieve control objectives and accomplish challenging tasks.