An approach for validating safety of perception software in autonomous driving systems

Abstract

The increasing complexity of Advanced Driver Assistance Systems (ADAS) presents a challenging task to validate safe and reliable performance of these systems under varied conditions. The test and validation of ADAS/AD with real test drives, although important, involves huge costs and time. Simulation tools provide an alternative with the added advantage of reproducibility but often use ideal sensors, which do not reflect real sensor output accurately. This paper presents a new validation methodology using fault injection, as recommended by the ISO 26262 standard, to test software and system robustness. In our work, we investigated and developed a tool capable of inserting faults at different software and system levels to verify its robustness. The scope of this paper is to cover the fault injection test for the Visteon’s DriveCore™ system, a centralized domain controller for Autonomous driving which is sensor agnostic and SoC agnostic. With this new approach, the validation of safety monitoring functionality and its behavior can be tested using real-world data instead of synthetic data from simulation tools resulting in having better confidence in system performance before proceeding with in-vehicle testing.