Grid based Estimation of Decision Uncertainty of Autonomous Driving Systems using Belief Function theory

Abstract

Autonomous driving systems take decisions under uncertainty present in the environment perception and localization due to the noise in the sensor measurements. This input uncertainty will be propagated to the decisions in some form. In this paper, we define the decision uncertainty in terms of degree of belief in the admissibility of the driving command. It is related to the uncertainty in the environment perception using belief function theory with a subset of discrete cells of the occupancy grid map as the source of evidences. We define an evidential framework and a mass assignment function for the propositions of interest. The evidences are combined using Dempster’s combination rule and Weighted Average method. This decision uncertainty can be seen as the level of confidence in the decision with respect to its admissibility. It can be displayed in the feedback to the human driver during the autonomous mode or can be used for giving the preemptive collision warning or fed back to the autonomous system itself to improve its performance. It can also be used for the division of control authority in the case of shared driving. The method is validated in the simulated environment of SCANeR Studio and the results obtained using the two combination methods are compared.