Concretize: A Model-Driven Tool for Scenario-Based Autonomous Vehicle Testing

To achieve rigorous certification of autonomous vehicles (AVs), testing approaches must handle all possible, practically relevant traffic scenarios. This is achievable through the handling of relevant abstractions within the scenario specification language and throughout the scenario generation process. While many scenario generation approaches exist, they are often limited to generating instances of a fixed (pre-defined) scenario and lack tool support. In this paper, we introduce Concretize, a model-driven AV testing framework. It (1) allows users to define scenario specifications using an abstract domain-specific language, and (2) generates conforming concrete (exact) scenarios, which are (3) visualized via a user-friendly web interface. Scenarios are also (4) executed in simulation, in which case Concretize (5) auto-generates figures depicting the monitored safety behavior of the AV-under-test wrt. scenario components at various abstraction levels.