Operational Domain Metamodel for Testing AI Systems in Aviation

摘要

View Video Presentation: https://doi.org/10.2514/6.2023-2589.vid Scenario-based testing in simulation is replacing real-life testing for autonomous systems. The main components of a scenario model are environmental factors, the entities and the events. One of the significant aspects of the scenarios is the representation of the system's dynamic behaviour under test. Many languages have been working on modelling the dynamic content of the scenarios in the Automated Vehicles domain. One of the concepts used is a Behaviour Tree, used in the gaming and robotics industry to model the AI behaviour of the system. This paper introduces a new approach to scenario-based testing of aviation systems called Operational Domain Driven Testing. It enhances the existing scenario process using an ontology called System Entity Structures with Behaviour trees. The ontology is suitable for variable structural modelling of the scenario's entities and environment, whereas the Behaviour Trees handle the dynamic nature of events. The authors demonstrate the approach using a use case to validate the Drone's perception algorithms in an urban environment with successful execution in the Gazebo simulator.