Computing risk for Unmanned Aircraft self separation with maneuvering intruders

Safe integration of Unmanned Aircraft Systems (UAS) into the civil airspace requires the development of a sense and avoid (SAA) capability that enables UAS to remain "well clear" from other airborne traffic. Providing this capability when encountering non-cooperative, maneuvering intruders, such as those operating under Visual Flight Rules (VFR), is particularly challenging due to the inherent uncertainties in predicting the future trajectories of these intruders. Experts have suggested [1] that one way of meeting this challenge is to treat "well clear" as a separation standard that is quantified using the risk (i.e. probability) of Near Mid-Air Collision (NMAC) at some future time, and to alert pilots when action is required to avoid violating this separation. This involves (explicitly or implicitly) a stochastic model to quantify likely intruder trajectories. In this paper, we develop algorithmic tools for computing such risk by expanding techniques developed in the target tracking community. A central feature of this approach is the use of continuous-time, maneuver-based (rather than traditional diffusion-based) stochastic models that are more representative of variations in maneuvering aircraft trajectories over longer time scales. We argue that evaluating risk using such models is computationally viable for a real-time SAA system and can provide enhanced performance in terms of the traditional detection-theoretic metrics of probability of detection (Pd) and probability of false alarm (Pfa).