Research on the SSIDM Modeling Mechanism for Equivalent Driver’s Behavior

Abstract

In order to address the challenge of seamlessly transitioning between the car-following model and lane-changing model, we employed the Intelligent Driver Model (IDM) within a single lane context to investigate how drivers switch their behavior. This involved studying their actions when following a vehicle normally, deciding to change lanes, creating space and gaining speed for the lane change, and executing the actual lane change. When there was sufficient room for lane-changing and an opportunity for speed gains, we considered the ego vehicle's intention to change lanes as the solution to the boundary between car-following and lane-changing behaviors, a point often characterized as the IDM's failure threshold. However, in cases where there were no significant advantages to changing lanes, we improved the IDM by integrating the constraints posed by nearby vehicles in the target lane along with the ego vehicle's current state. This refinement resulted in the creation of the Stepless Switching Intelligent Driver Model (SSIDM). To gather real-world driving data, we collected natural driving information from actual drivers and conducted scenario analysis on structured road networks. Utilizing this collected dataset, we employed an elliptic equation to accurately define the boundary for switching between these driving behaviors. Additionally, we determined the optimal balance coefficients for the influence of leading and trailing vehicles within the target lane. The verification results of the test set indicated that the Mean Square Error (MSE) of the SSIDM was 2.172, which represents a substantial 57.98% reduction compared to the conventional single-lane IDM. The SSIDM can smoothly and precisely replicate the transition between car-following and lane-changing behaviors, demonstrating greater accuracy than the IDM. This research not only offers valuable theoretical support for developing point-to-point driving models but also contributes to the advancement of L2+ autonomous driving capabilities. It has the potential to facilitate the practical implementation of comprehensive behavior and scene-aware autonomous driving systems.