Optimization of a lateral driver assistance function by combining classical approaches and artificial intelligence

Authors

C. Olbrich, L. Witt, D. Münning - Volkswagen AG, D. Engel - HAW Hamburg

Summary

This paper investigates the use of artificial intelligence in the field of advanced driver assistance systems. A frequently used method for sequential decision problems is reinforcement learning. Advantage of this method is the experience-based learning of the model in its environment. The system considered is a conventional vehicle controller that guides the vehicle in the lateral direction. Aim of the concept is to use a reinforcement learning algorithm to learn the calibration parameters of a conventional controller in order to reduce the effort of experimental or manual parameter calibration, enable a comfortable driving behavior and also react to environmental or vehicle changes. In this article, an initial overview of advanced driver assistance systems and artificial intelligence is provided. Furthermore, related work on this topic is presented, followed by a description and potential implementations of the developed concept. Finally, there is a concluding discussion about the feasibility of the chosen concept.