Typical-driving-style-oriented Personalized Adaptive Cruise Control design based on human driving data

摘要

Reflecting different driving styles in Adaptive Cruise Control (ACC) is of great importance for its market acceptance. A novel data-based method is presented for designing a Personalized Adaptive Cruise Control (PACC) system in this paper. First, a driving-data-acquisition platform is established, and a large amount of real-world driving data from 84 human drivers is collected. To measure the similarity of human drivers quantitatively, the driving data of every driver are regarded as a specific distribution of some features, fitted with a Gaussian mixture model (GMM). Kullback-Leibler (KL) divergence is introduced as the driving similarity index. After that, an unsupervised clustering algorithm is realized in this paper, and these drivers are grouped into three separate groups. A practical PACC structure is designed in the second stage based on the grouped driving data to include different driving characteristics, mainly in three aspects: speed control, distance control, and the switching rule. Then real-vehicle experiments are carried out. Results demonstrate the capabilities of the proposed PACC algorithm to reflect different driving styles.