Falsification of a Vision-based Automatic Landing System

TL;DR

This paper investigates the vulnerability of a vision-based automatic landing system for aircraft by applying falsification techniques to identify potential safety violations, using simulation data and formal specifications.

Contribution

It introduces a formal framework for falsifying vision-based landing systems and demonstrates its effectiveness through simulation-based experiments.

Findings

Falsification can identify safety violations in the landing system.

Simulation experiments validate the approach's ability to find counterexamples.

The method enhances safety analysis for vision-based aircraft landing systems.

Abstract

At smaller airports without an instrument approach or advanced equipment, automatic landing of aircraft is a safety-critical task that requires the use of sensors present on the aircraft. In this paper, we study falsification of an automatic landing system for fixed-wing aircraft using a camera as its main sensor. We first present an architecture for vision-based automatic landing, including a vision-based runway distance and orientation estimator and an associated PID controller. We then outline landing specifications that we validate with actual flight data. Using these specifications, we propose the use of the falsification tool Breach to find counterexamples to the specifications in the automatic landing system. Our experiments are implemented using a Beechcraft Baron 58 in the X-Plane flight simulator communicating with MATLAB Simulink.