Self-driving cars pose significant testing challenges. The quantity and magnitude of sensor inputs create an enormous test-vector space, making it impractical for traditional manual testing to provide adequate coverage.
The article proposed model-based techniques to manage this complexity. Abstraction and automation can increase test-engineering productivity by supporting the generation of:
- Test cases
- Test data
- Predicted system behavior
The described strategy included sensor test generation, automated navigation testing, and turbulence-induced testing. The original paper presented a case study and summarized conclusions from applying these techniques.
*Historical note: this article and paper date from August 2017. They describe an earlier stage of autonomous-vehicle testing; they should be read as a case study rather than a survey of current industry practice.*
