Automated Probabilistic Analysis of Air Traffic Control Communications

Abstract

: Initiatives to integrate autonomous Unmanned Aerial Vehicles (UAVs) with regular airport operations require automated onboard situational awareness to maintain safety at all times. More specifically, this requires the capability to sense, interpret, and predict what other aircraft are doing, based on the same incoming data that are available to a human pilot. This includes not only baseline knowledge of the airport layout, operational practices and landmarks, but also an ability to interpret radio communications with Air Traffic Control (ATC) and correlate them with observable movements and positions of other aircraft. This analysis informs an autonomous UAV's control mechanisms which ultimately regulate its kinetic behavior at the airport. As with any operational domain governed by human actions and control, there are many inherent challenges in interpreting ATC communications -- a noisy data stream not only in terms of signal quality, but more significantly in the range of human deviations from the strictest procedures. This makes the analysis a natural application for Artificial Intelligence techniques, where the goal is to support automated reasoning that mimics a human pilot's decision processes. This paper provides a detailed discussion of a probabilistic reasoning approach using Bayesian Networks to classify ATC communications and synthesize them with baseline knowledge of an airport and produce real-time hypotheses about the states and trajectories of other aircraft. This provides a key component for automated situational awareness, which also requires correlation with sensor data, and ultimately a functional set of behaviors to act accordingly, although these latter capabilities are beyond the scope of this paper. The probabilistic communications analysis methodology is described, along with testing results using a real-world sample data set annotated for ground truth, to evaluate performance.