Stuart Reid - Chairman, Software Testing Specialist Group
The testing of traditional (deterministic) safety-related systems is well-understood, but the advent of autonomous vehicles has introduced the need for Artificial Intelligence in the form of self-learning systems.
These systems are creating new challenges for testers, both for the vendors of autonomous systems and for the regulators. This presentation explores four major challenges. The process of creating a self-learning system opens up new opportunities for errors and defects, which testers and quality assurers must both consider.
The complexity of self-learning systems has meant that most testing has been black-box, but their criticality when used for autonomous vehicles means that white-box testing must also now be considered.
The difficulties of assuring these systems and the balance of costs between development and testing means that developers must consider the architectures needed to make the testing and assurance possible.
Finally, the costs inherent in testing probabilistic systems will mean that virtual testing will have to play a far more prominent role.