Testing Method for Multi-UAV Conflict Resolution Using Agent-Based Simulation and Multi-Objective Search

A new approach to testing multi-UAV conflict resolution algorithms is presented. The problem is formulated as a multi-objective search problem with two objectives: finding air traffic encounters that 1) are able to reveal faults in conflict resolution algorithms and 2) are likely to happen in the real world. The method uses agent-based simulation and multi-objective search to automatically find encounters satisfying these objectives. It describes pairwise encounters in three-dimensional space using a parameterized geometry representation, which allows encounters involving multiple UAVs to be generated by combining several pairwise encounters. The consequences of the encounters, given the conflict resolution algorithm, are explored using a fast-time agent-based simulator. To find encounters meeting the two objectives, a genetic algorithm approach is used. The method is applied to test ORCA-3D, a widely cited open-source multi-UAV conflict resolution algorithm, and the method’s performance is compared with a plausible random testing approach. The results show that the method can find the required encounters more efficiently than the random search. The identified safety incidents are then the starting points for understanding limitations of the conflict resolution algorithm.