The Curse of the Time Horizon in Detect & Avoid Algorithms

This paper introduces a centralized collision avoidance algorithm based on a Differential Evolution, for the purpose of studying the time horizon effect, a pathological behavior previously identified in Detect & Avoid (D&A) decentralized algorithms based on geometric methods. This pathological behavior, called time horizon effect, is most likely to occur during constant-speed encounters, when the lateral maneuvers issued by the D&A system postpone the crossing of trajectories beyond the horizon of conflict detection by maneuvering the flights toward parallel tracks. In such cases, the flights might end up locked on parallel tracks, missing their destination. The proposed centralized algorithm selects the best direction changes at each time step for all flights, with the primary objective to maintain a minimum separation between the flights. We show that the time horizon effect also occurs when using such a centralized optimization algorithm having full knowledge of the flight intents, and propose mitigating strategies. This suggests that the horizon effect is not related to the distributed nature of many D&A methods found in the literature, but is rather a more general effect due to the myopic nature of the decision process.