Interaction-Aware Approach for Online Parameter Estimation of a Multi-lane Intelligent Driver Model

The paper presents a probabilistic approach for online parameter estimation of an enhanced car-following model appropriate for multi-lane traffic, which is based on an extension of the well-known Intelligent Driver Model (IDM). The approach explicitly considers the simultaneous influence of several interacting vehicles on the longitudinal dynamics of the ego-vehicle. Therefore, a method to extract the relevant reference vehicles considered in the proposed multi-lane car-following model is developed. In order to calibrate the model parameters online, a particle filter approach, which is able to deal with the overdetermined model structure, is employed. Experimental studies using a real highway scenario observed by vehicle surroundings sensors show the need of the online-calibrated multi-lane architecture. To use the model for vehicle speed prediction, a further learning-based extension is suggested which enables the adaption of the model parameters over the prediction horizon.