With the rapid development of intelligent transportation system and autonomous driving technology, Adaptive Cruise Control (ACC) system has become a key component of modern intelligent vehicles. However, the traditional rule-based ACC system shows the problem of insufficient flexibility in complex and changeable traffic environment. Therefore, a new ACC algorithm is designed by introducing reinforcement learning (RL) technology, especially hierarchical RL framework, in order to improve the adaptability and anthropomorphic driving style of the system. Firstly, this paper constructs a hierarchical RL framework, which includes high-level policy network (HPLN) and low-level execution network (LLEN). Secondly, in order to learn anthropomorphic driving strategy, this paper adopts the inverse RL (IRL) method. By collecting the driving data of human drivers in different road environments and traffic conditions, IRL algorithm is used to learn a reward function that can reflect human driving styles and preferences. Then, Proximal Policy Optimization (PPO) algorithm is used to train the strategy network to maximize the learned reward function, so as to realize anthropomorphic driving behavior. This study is verified by experiments on the CARLA simulation platform. The experimental results show that the algorithm has obvious advantages in safety, comfort and following, and can effectively adapt to the complex and changeable traffic environment and realize a more anthropomorphic driving style. Compared with other comparative algorithms, the algorithm proposed in this paper shows superior performance in key indexes such as collision times, sudden braking times, acceleration change rate and bump degree. The development method of automobile ACC algorithm based on RL proposed in this paper effectively improves the adaptive ability and anthropomorphic driving style of ACC system by introducing hierarchical RL framework and inverse RL technology, and provides new ideas and methods for the development of intelligent driving technology.
Development of Automotive Adaptive Cruise Control Algorithm Based on Reinforcement Learning
2024-08-14
340845 byte
Conference paper
Electronic Resource
English
INVERSE REINFORCEMENT LEARNING FOR ADAPTIVE CRUISE CONTROL
European Patent Office | 2024
|Multi-Objective Adaptive Cruise Control via Deep Reinforcement Learning
British Library Conference Proceedings | 2022
|Multi-Objective Adaptive Cruise Control via Deep Reinforcement Learning
British Library Conference Proceedings | 2022
|Multi-Objective Adaptive Cruise Control via Deep Reinforcement Learning
British Library Conference Proceedings | 2022
|