Behaviour Tree Visualisation for Autonomous Aircraft Simulation

Sammanfattning

Simulation is often used when developing autonomous aircraft, since it makes it possible to run the same scenario several times and inspect parts of the system that would be difficult to see during real flight. However, it can be hard for engineers to understand why an autonomous aircraft behaved in a certain way, especially when relevant information is spread between logs, spatial views, internal states, and decision logic. This thesis investigates how visualisation can support analysis of internal autonomy processes in autonomous aircraft simulation. The work was done in three steps: two pre-study group interviews with engineers, development of a visualisation prototype, and a qualitative evaluation using simulation scenarios. The prototype was built with ROS2, Gazebo, PX4, RViz, and Behaviour Trees. It combines a spatial view of the aircraft and mission with visualisation of Behaviour Tree execution and selected internal values. The results suggest that the most useful support came from combining several views rather than from one single visual element. In the evaluation, the participants used the spatial view, Behaviour Tree viewer, and Behaviour Tree heads-up display together when reasoning about both normal and anomalous behaviour. Without the internal visualisation, they could still make reasonable guesses about possible errors, but had less support for checking those guesses. The study is exploratory and limited, but indicates that visualising internal autonomy states can make simulation-based analysis more grounded and less dependent on unsupported assumptions.