Assumptions of Lateral Acceleration Behavior Limits for Prediction Tasks in Autonomous Vehicles

This paper presents an analysis of the euroFot data set to determine limits for the typical lateral acceleration behavior of drivers. Since recent studies indicate that lateral accelerations close to the physically possible limit are rarely used by drivers, predictions tasks for autonomous driving could consider a smaller, so-called natural lateral acceleration interval (NLAI) instead of all physically possible lateral accelerations. This NLAI should be as small as possible while still fulfilling all safety aspects. Therefore, valid assumptions are required on which the interval can be derived. Since a valid assumption which leads to minimal NLAI is yet unknown, four different assumptions concerning the lateral acceleration behavior are derived and evaluated in this paper. Thereby, detailed examinations regarding the relative frequencies of violations are presented. Finally, two assumptions are recommended for introducing an NLAI, depending on prediction time and safety requirements. Additionally, the advantages of utilizing an NLAI instead of all physically possible lateral accelerations are highlighted by comparing the results of an occupancy prediction approach.