Analyzing mandatory and discretionary lane change interaction patterns using hidden Markov model-based approaches

It is indispensable for autonomous vehicles (AVs) to understand the complex and dynamic lane-changing interaction patterns, which can support AVs in making appropriate driving decisions. This study proposed a learning framework for understanding the interaction patterns during mandatory lane change (MLC) and discretionary lane change (DLC). Three hidden Markov model (HMM) based approaches, namely HMM with Gaussian mixture model (GMM-HMM), hierarchical Dirichlet process-hidden semi-Markov model (HDP-HSMM), and coupled HMM (CHMM) are compared for segmenting driving primitives. Then dynamic time warping distance-based K-means clustering is employed to group the driving primitives into 6 and 8 interaction patterns for MLC and DLC. The minimum time to collision (TTC) of two conflict types between interactive vehicles involved in the lane-changing scenario is applied to evaluate the traffic risk associated with interaction patterns. Two types of lane-changing events are extracted at a freeway entrance ramp from the international, adversarial, and cooperative motion (INTERACTION) dataset. The experimental results demonstrate that the HDP-HSMM achieves better performance in separating driving primitives with interpretable semantic information, enabling a comprehensive understanding of the dynamic spatiotemporal characteristics and the traffic risk evolution mechanisms of lane-changing interaction patterns. Additionally, the traffic risk associated with interaction patterns of DLC is generally higher than that of MLC. The findings of this study are beneficial for AVs in understanding the collision risk during lane changes, thereby enhancing driving decision-making.