Autonomous Driving Decision Model based on Reinforcement Learning Algorithm in the Environment of Intelligent Connected Vehicles

Engineering notes

Key topics: autonomous driving, reinforcement learning, carla. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

With the rapid development of autonomous driving technology, how to achieve safe, efficient and comfortable vehicle autonomous decision-making in complex traffic environments has become a key challenge. The proximal policy optimization algorithm (PPO) in reinforcement learning has been widely used in the design of autonomous driving decision-making systems due to its stable training process and good performance.Based on the PPO algorithm, this study designed a state space and multi-objective reward function that adapts to different driving scenarios, covering urban roads, highways and human-computer interaction scenarios. A large number of experiments were conducted on the CARLA, SUMO and TORCS simulation platforms to simulate real traffic environments, and the safety, efficiency and driving comfort of the algorithm were systematically evaluated.The experimental results show that the PPO algorithm can achieve high safety in a variety of complex scenarios, with the average vehicle speed on urban roads maintained between 28.4 and 34 km/h, the success rate of overtaking on highways reaching 72.3% to 90.5%, and the number of emergency brakes effectively reduced. Despite this, the algorithm is still unstable under extremely complex traffic conditions, showing the potential for further improvement.

Links and sources

• Official / arXiv page

Comments