Smart Testing of Intelligent Systems

At present, engineers are facing a huge challenge

when trying to effectively develop test cases as part

of the validation process in the area of highly

automated driving. The number of tests required for

validating safety-critical ECU functionalities in this

field is increasing drastically. Therefore, countless

driving situations and scenarios are required to

ensure a proper validation. It has already become

impossible to validate this exponentially growing task

reliably with prototype and real vehicles alone. The

relevant tests must be moved to the lab earlier in the

development process by the means of simulation.

However, the well-known and established method of

requirements-based testing does not seem to be

adequate as it relies on existing straight-forward test

catalogs. For a sufficient test coverage, it is therefore

necessary to complement the standard method with

additional approaches.

This paper focuses on a new approach that quickly

reveals violations of various acceptance criteria.

Simultaneously, it guarantees a reliable coverage of

relevant traffic situations. Mathematical algorithms,

based on test result histories, automatically generate

test parameters for critical situations. To support

flexible development and testing of functions of

autonomous driving, this framework allows end users

to integrate their own algorithms and scenarios into

the validation process.