Towards Realizing Test Conditions for Automated Vehicles

The verification and validation of Automated Driving Systems pose a challenge to the deployment of automated vehicles onto the roads. Current research considers a scenario-based approach to control the traffic situations an automated vehicle is tested in. One of the several open challenges in this field is how test conditions for the self-driving vehicle under test (VUT) can be realized. This is because the test designer probably cannot tailor the test script to the VUT's actual behaviour due to unknown knowledge of the complex system logic and therefore the VUT might not reach the situation of interest during a test. In this paper, the problem of bringing the VUT and the environment in the desired initial states for a test, respectively a scenario, is described. Furthermore, the idea of a reinforcement learning approach of controlling traffic agents in the environment of the VUT is presented and under development in the corresponding doctoral project with focus on virtual testing.