LM-IDM: A novel multi-lane car-following model considering lateral merging in mixed traffic

With the advancement of Autonomous Vehicles (AVs), modeling their interaction with Human-Driven Vehicles (HDVs) in complex multi-lane scenarios, such as ramp merges, has become a critical challenge. Traditional car-following models often neglect the coupling of lateral and longitudinal dynamics, thus failing to capture realistic driving behaviors. To address this gap, this paper proposes a novel multi-lane car-following model: the Lateral-Merge Intelligent Driver Model (LM-IDM). This model integrates the principles of the Social Force Model and the Intelligent Driver Model, leveraging two-dimensional perception to dynamically simulate the influence of surrounding vehicles' lateral positions on the ego vehicle's longitudinal acceleration, thereby enabling more human-like responses to merging events. Microscopic validation on the CitySim and exiD datasets confirms the model's superiority: LM-IDM significantly outperforms traditional and data-driven benchmarks in trajectory accuracy and behavioral realism, achieving robust safety in scenarios where the LSTM model failed catastrophically. Notably, these performance advantages are achieved alongside substantial improvements in driving comfort, fuel economy, and exceptional computational efficiency. Macroscopic simulations further reveal that a high penetration of LM-IDM-equipped vehicles can significantly enhance traffic flow efficiency and stability at bottlenecks. These findings establish LM-IDM as a robust, efficient, and transferable model, underscoring the value of physically-grounded principles in developing safe and reliable autonomous driving solutions.