A Learning Approach to Air Traffic Control

摘要

Reinforcement learning, as a means of the general universal trend towards adaptation, is considered a suitable methodology for air traffic control. MAGENTA is a generalized framework for this kind of reinforcement learning. MAGENTA can be applied introducing modular learning, using a pathfinder that tries to reach a target, first in an area containing static objects, then in another containing dynamically rolling balls. In the second phase, simulation then helps the pathfinder/aircraft to learn by taking a direction and heading towards the start of the runway, staying within the virtual conical space for proper approach and landing, and avoiding collision with other aircraft in the area whose target is the same and that are heading towards the start of the same airport runway.