Geometric Verification of Vector Fields for Autonomous Aircraft Navigation

Abstract

New generations of small advanced aircraft may soon transform how people and freight are transported. Thus, ensuring safe and reliable operation of these systems, in all plausible scenarios, is of utmost priority. In this paper, we address this problem by proposing a method to verify the safety of an autonomous aircraft controller following a navigational policy. The policy is encoded using discrete vector fields that are meant to drive the vehicle to a goal state and away from obstacles or restricted airspace. The solution presented in this paper provides the ability to verify the policy, catching unsafe scenarios which may be missed by random Monte Carlo methods or general reachability analysis. After illustrating the main theoretical principles, a possible practical implementation is described and compared with analysis based on Monte Carlo simulations. The comparison shows an insightful example where Monte Carlo simulations fail to detect several corner cases that are uncovered by the proposed method. In the conclusion of the paper, we provide insights about possible future research to implement the proposed solution obtaining higher accuracy and efficiency. INDEX TERMS Navigation, Unmanned aerial vehicles, Autonomous agents, Reachability analysis VOLUME 4, 2016 This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/ACCESS.2021.3062662, IEEE Access Giovanni Miraglia et al.: Geometric Verification of Vector Fields for Autonomous Aircraft Navigation Giovanni Miraglia et al.: Geometric Verification of Vector Fields for Autonomous Aircraft Navigation LOYD R. HOOK (Member, IEEE) has worked in aircraft autonomous systems research and implementation for the past 15 years beginning at NASA Dryden/Armstrong flight research center and continuing at his current position as Assistant Professor at the University of Tulsa. Loyd is currently the director of the TU Vehicle Autonomy and Intelligence Lab (TU VAIL) where his research focus is the development of safety assured autonomous vehicle systems. Loyd has MS and Ph.D. degrees from the University of Oklahoma in Electrical and Computer Engineering from 2006 and 2011 respectively.