Conflict resolution in a decentralized air traffic concept of operation

Abstract

The current air traffic concept of operations relies on a centralized process in which ground controllers are responsible for determining conflict-free trajectories. However, with new technologies such as ADS-B and GPS, aircraft could directly interact together to resolve their own conflicts in a decentralized manner. The challenge is to guarantee aircraft separation while converging to reasonably fair resolutions for all aircraft. The difficulty is that aircraft have only limited information about how the other aircraft evaluate the cost of conflict resolutions. Thus, this thesis proposes to frame decentralized conflict resolution using game theory. A collaborative decentralized conflict resolution is developed as a sequential bargaining process between the different aircraft. The goal of each aircraft is to minimize the cost associated with the conflict resolution. However, each aircraft doesn’t know the cost function and performance constraints of the other involved aircraft. In the sequential bargaining process developed, aircraft propose at each step personal trajectories to the other aircraft, corresponding to trajectories they would be ready to fly. Then they compute response trajectories, corresponding to trajectories they would have to fly to avoid the conflict if the personal trajectories were flown. If some response trajectories are cheaper than the offered personal trajectories, an agreement is reached; otherwise compromises have to be made by the aircraft by offering more expensive personal trajectories at the next step. Several pairwise conflict experiments, corresponding to different conflict geometries, were conducted to explore different ways of handling performance constraints and different ways of searching trajectories in the resolution space. Ultimately, the algorithm was demonstrated in a large scale simulation with more than a thousand aircraft flying over the Indianapolis Center, incurring more than five hundred conflicts. The traffic sets were taken from real ETMS data over five hours, to represent ‘real’ conditions. 93% of the conflicts were successfully solved by the bargaining process.