Application of reinforcement learning to detect and mitigate airspace loss of separation events

Abstract

The volume of both manned and unmanned air traffic in the National Airspace (NAS) is projected to increase substantially over the coming decades with the consequence of increasing Air Traffic Control (ATC) workload, airspace congestion and the risk of mid-air collisions. Current ATC traffic management practices are human intensive. Separation is managed by ATC through open-loop vectoring and monitored on-board through collision avoidance systems such as the Traffic Collision Avoidance System (TCAS). In this paper, we discuss a machine learning based system that uses real-time system-wide traffic surveillance data to identify anomalous traffic behaviors that can lead to loss of separation (LOS) events. Specifically, this work presents an application of reinforcement learning to detect and mitigate impending airspace loss of separation events. We discuss the model representation and learning techniques, demonstrate the alert and recommended model actions, review our findings, and highlight future steps. With the mandatory Automatic Dependent Surveillance-Broadcast (ADS-B) usage being enforced in the NAS by 2020, it is expected that a significant amount of real-time traffic surveillance data will be available to leverage and build upon the developed technique.