Mining Test Inputs for Autonomous Vehicles

Abstract

Testing of Advanced Driving Assistance Systems (ADAS) with higher level of autonomy is becoming increasingly complex. A variety of situations needs to be tested to ensure a sufficient test coverage. In normal driving condition, this requires enormous amounts of driven kilometers. Our presented approach extracts scenario information from observed simulations and identifies test-cases.

In order to test such system it is necessary to generate test-cases automatically and to execute the test-cases using simulators. The presented approach uses observed behaviors to mine partitions of the test input automatically. The test input consists of continuous behavior of other traffic participants which is supposed to trigger behaviors of the tested vehicle. We discuss how to identify cause-effect relationships and how these relationships can be assembled to test-cases. We demonstrate our approach by analyzing a lane change scenarios of trucks on highways.