Incremental Reliability Modeling Framework to Optimize Vehicle's Validation Tests

This paper deals with the problem of the reliability certification for an autonomous driving (AD) system and provides some insights for assessing the system reliability in a minimum testing period. Preliminary works [1] propose a decomposition of the system reliability as a function of known and unknown driving scenarios. Here, the method focuses on the reliability assessment when the driving situations are assumed to be identified and partially known. A Bayesian framework is designed for mixing numerical simulation based and field data obtained during open road tests. Moreover, for reducing the number of kilometers in the open road tests, a selection of the “most critical” scenarios should be done. A simplified real-world example has been designed to illustrate the efficiency of the methodology in terms of speed of convergence for the validation process.