Human-like car-following modeling based on online driving style recognition

Abstract

Incorporating human driving style into car-following modeling is critical for achieving higher levels of driving automation. By capturing the characteristics of human driving, it can lead to a more natural and seamless transition from human-driven to automated driving. A clustering approach is introduced that utilized principal component analysis (PCA) and k-means clustering algorithm to identify driving style types such as aggressive, moderate and conservative at the timestep level. Additionally, an online driving style recognition technique is developed based on the memory effect in driving behavior, allowing for real-time identification of a driver's driving style and enabling adaptive control in automated driving. Finally, the Intelligent Driver Model (IDM) has been improved through the incorporation of an online driving style recognition strategy into car-following modeling, resulting in a human-like IDM that emulates real-world driving behaviors. This enhancement has important implications for the field of automated driving, as it allows for greater accuracy and adaptability in modeling human driving behavior and may ultimately lead to more effective and seamless transitions between human-driven and automated driving modes. The results show that the time-step level driving style recognition method provides a more precise understanding of driving styles that accounts for both inter-driver heterogeneity and intra-driver variation. The proposed human-like IDM performs well in capturing driving style characteristics and reproducing driving behavior. The stability of this improved human-like IDM is also confirmed, indicating its reliability and effectiveness. Overall, the research suggests that the proposed model has promising performance and potential applications in the field of automated driving.