Applying Decision Graphs in the Context of Automated Driving

Abstract

Techniques to enable automated driving currently receive a lot of attention in computer science research. Car automation requires realizing several cognitive functions by computers. One important functionality is environment perception. This consists of several sub-tasks which are complex and thus computation intensive when implemented. We propose the use of decision graphs to speed up the execution of a consistency check. This check is applied to the output of a of neural net which classifies regions in an environment image. The check consists in evaluating a set of probabilistic rules. The paper describes how the miAamics approach of pre-computing results of rule evaluations with decision graphs may be profitably used in this application.