Behavior planning for automated highway driving

Abstract

This work deals with certain components of an automated driving system for highways, focusing on lane change behavior planning. It presents a variety of algorithms of a modular system aiming at safe and comfortable driving. A major contribution of this work is a method for analyzing traffic scenes in a spatio-temporal, curvilinear coordinate system. The results of this analysis are used in a further step to generate lane change trajectories. A total of three approaches with increasing levels of complexity and capabilities are compared. The most advanced approach formulates the problem as a linear-quadratic cooperative game and accounts for the inherently uncertain and multimodal nature of trajectory predictions for surrounding road users. Evaluations on real data show that the developed algorithms can be integrated into current generation automated driving software systems fulfilling runtime constraints.