Project Summary

Automated generation of critical test cases for CAV (connected automated vehicle) verification & validation

Project Achievements

We have used a realistic and reasonably complex automated driving scenario, an ACC (Adaptive Cruise Control) system responding to a cut out driving manoeuvre, as a case study to develop our algorithms. After identifying the key variations to this scenario and the key performance indicators to measure this scenario, we investigated a range of machine learning algorithms to carry out the active search and to cluster the results.

Conclusions

We have shown promising results in finding issues in our device-under-test. After only 20 iterations of an example optimisation algorithm, we have increased the proportion of issues found by 100x.

After finding these issues, we have clustered them together into common-cause groups, reducing the number of items reported to the development engineer by 50x so that it is easier for them to understand and therefore to resolve.