This talk will present our efforts to advance decision making in the Air Traffic Management (ATM) domain, using Reinforcement Learning methods. Specifically, it will explain ATM terminology and concerns, and will show our efforts on devising scalable multi agent reinforcement learning methods towards resolving problems in this complex domain. In doing so, collaborative methods, as well as hierarchical reinforcement methods enforcing state and action abstractions are studied under a common framework of challenging problems. Future prospects towards the application of multi agent reinforcement learning methods will also be examined.
A EURAMAS-sponsored tutorial.