Real-Time Mitigation of Loss of Separation Events using Reinforcement Learning

With the projected increase of manned and unmanned air traffic in the National Airspace (NAS), an uptick in airspace congestion and the risk of mid-air collisions is inevitable. Current collision avoidance systems such as the Traffic Alert and Collision Avoidance System (TCAS) and the Automatic Dependent Surveillance - Broadcast (ADS-B) functions provide an aircraft-centric view and do not account for complex traffic patterns in congested airspace. This projected increase in air traffic will in turn place undue burden on the separation services provided by Air Traffic Control (ATC). To ease the burden on ATC, we propose a reinforcement learning framework that considers traffic patterns and provides ATC with i) real-time alerts on impending potential loss of separation events and ii) a suggestive course of action to mitigate loss of separation. At runtime, the technique informs ATC of the best course of action to take to mitigate loss of separation within a terminal area using real-time system-wide traffic surveillance data. With the mandatory ADS-B usage being enforced in the NAS by 2020, a significant amount of real-time traffic surveillance data will be available to leverage the developed technique. Our primary contribution is the development of a reinforcement learning framework to predict and mitigate potential loss of separation events in congested airspaces with intersecting arrival and departure paths. We present results from the application of the proposed approach using data from the New York Metroplex airspace.