Exploring Model Quality for ACAS X

Abstract

The next generation airborne collision avoidance system, ACAS X, aims to provide robustness through a probabilistic model that represents sources of uncertainty. From this model, dynamic programming produces a look-up table that is used to give advisories to the pilot in real time. The model is not present in the final system and is therefore not included in the standard certification processes. Rather, the model is checked indirectly, by ensuring that ACAS X performs as well as, or better than, the state-of-the-art, TCAS. We claim that to build confidence in such systems, it is important to target model quality directly. We investigate this issue of model quality as part of our research on informing certification standards for autonomy. Using ACAS X as our driving example, we study the relationship between the probabilistic model and the real world, in an attempt to characterize the quality of the model for the purpose of building ACAS X. This paper presents model conformance metrics, their application to ACAS X, and the results that we obtained from our study.

Three-dimensional (3D) Monte-Carlo modeling for UAS collision risk management in restricted airport airspace

“…Both methods could be enhanced with state-update if a new position report for the intruder becomes available [15,25,26]. Modern probabilistic path prediction further enhances its accuracy using machine learning that takes advantage of the massive air traffic track data that is available to predict the intent of the intruder [27][28][29][30][31][32][33].…”