Towards Measuring the Performance of Architectural Components of Autonomous Vehicular Systems

摘要

For a vehicular system to act 'intelligent', the system must be able to: (1) sense in a dynamic domain; (2) model the domain internally; (3) determine possible courses of action to accomplish a goal in the domain; and (4) be able to assess the various courses of actions to determine which is best. The actions that the system ultimately performs are a function of all of these components. Solely assigning performance metrics to the resultant action of the intelligent system does not evaluate any one of these components individually, and therefore leaves some doubt as to how to measure what each component contributes to the overall behavior of the system. Thus we are not looking at a single number, but a matrix of numbers that characterize the performance of the system. In this paper, we are exploring a mechanism to assign performance metrics to the part of the system that models the domain internally, the internal knowledge representation of intelligent vehicular systems. We do not consider that part of a system that translates the raw sensory input from a vehicle's sensors to other representations. Rather we simulate a predefined set of sensory inputs, and evaluate the resulting knowledge representation based.