A Modular Framework for Lateral Movement Modeling

The lateral movement of vehicles within their lane determines occurring occlusions, and thus decides on the range of visions of vehicle sensors used by automated driving functions and the objects detected by down-stream algorithms. As simulations play an integral role in the validation of automated driving functions, their ability to realistically model the lateral movement is crucial. However, currently applied methods such as microscopic traffic simulations and the scenario-based approach making use of maneuver-based scenario descriptions simplify or neglect the lateral movement of vehicles. For that reason, a two-level stochastic model has been introduced in earlier work. It consists of a Markov model for the systematic coarse movement and a noise model for the residual fine movement. In follow-up publications, several advancements for both model components have been presented. These follow a modular structure, thus, can be flexibly combined. This paper for the first time gives an overview over the whole modular framework for lateral movement modeling, and brings together the extensions proposed in individual publications. It further performs a systematic comparison of the possible model combinations. We analyze the implications of the various model configurations on model calibration, quantitative results, and runtime. The analyses reveal the extremes, namely the most accurate and the fastest model which are contrary to one another, but also solutions that lie in between. Based on these insights, one can make a well-founded decision on which modules to combine to best fulfill the requirements of a specific use case. This way the paper makes a contribution towards enhancing the realism of simulations adopted for the virtual and hybrid validation of automated driving functions and beyond.