Probabilistic Approach for Validation of Advanced Driver Assistance Systems

Abstract

This paper presents a methodological approach for validation of advanced driver assistance systems. The methodology relies on the use of randomized algorithms that are more efficient than conventional validation that uses simulations and field tests, especially with increasing complexity of the system. The methodology first specifies the perturbation space and performance criteria. Then, a minimum number of samples and a relevant sampling space are selected. Next, an iterative randomized simulation is executed; then the simulation model is validated with the use of hardware tests to increase the reliability of the estimated performance. The proof of concept is illustrated with some examples of a case study involving an adaptive cruise control system. The case study points out some characteristic properties of randomized algorithms with respect to the necessary sample complexity and sensitivity to model uncertainty. Solutions for these issues are proposed as are corresponding recommendations for research.