Longitudinal motion control algorithm for autonomous vehicles taking decisions based on the preceding vehicle behavior pattern

In the field of autonomous driving, a key concern is whether autonomous driving algorithms can better adapt to their environments. Currently, autonomous vehicles often adopt a single control strategy, which can reduce traffic efficiency and negatively impact other road users. To address this issue, this paper presents a longitudinal motion control algorithm for autonomous vehicles that makes decisions based on the preceding vehicle’s behavior pattern, aiming to comprehensively improve both traffic efficiency and safety. Firstly, using the NGSIM dataset, a large number of kinematic features from highway-driving vehicles are extracted and standardized. Subsequently, Principal Component Analysis (PCA) is applied to reduce dimensionality and decouple the data. Following this, Fuzzy C-Means clustering (FCM) is employed to categorize the vehicles’ driving characteristics into several typical behavior patterns. By incorporating traffic regulations from various countries, external metrics are established to evaluate the clustering results. Based on these metrics, the parameters are optimized to enhance the reliability of the clustering outcomes. Additionally, a preceding vehicle behavior pattern identification module was developed using a lightweight Convolutional Neural Network (CNN), achieving high accuracy and low computational load in online experiments. Depending on the different behavior patterns of the preceding vehicle, we design a safety distance model that balances safety and traffic efficiency. To ensure the target following distance is met, a longitudinal control module based on Deep Reinforcement Learning (DRL) is developed. Finally, comparative experiments are conducted, and the results demonstrate that the proposed algorithm effectively optimizes traffic efficiency, safety, and comfort in a comprehensive manner, thereby verifying its feasibility.