Title: Robust Navigation for Autonomous Fixed-wing Unmanned Aerial Vehicles
Author(s): Robert C. Leishman, Clark N. Taylor
Published in: Proceedings of the 30th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2017)
September 25 - 29, 2017
Oregon Convention Center
Portland, Oregon
Pages: 773 - 807
Cite this article: Leishman, Robert C., Taylor, Clark N., "Robust Navigation for Autonomous Fixed-wing Unmanned Aerial Vehicles," Proceedings of the 30th International Technical Meeting of The Satellite Division of the Institute of Navigation (ION GNSS+ 2017), Portland, Oregon, September 2017, pp. 773-807.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In
Abstract: The path forward for long-term, reliable aerial autonomy depends heavily on two tasks: the ability to correctly interpret information from sensors and compute reliable state estimates from those sensor measurements. Once reliable state estimates and measurement information are available, a vehicle can then complete autonomous behavior, such as computing control based on higher-level objectives and further processing/utilizing sensor data. However, without the ability to employ sensor measurements and estimate vital vehicle states, autonomy is not possible. With the advent of GPS, many autonomous or semi-autonomous behaviors have been enabled. However, when GPS information is unavailable or unreliable, particularly due to an adversarial environment, autonomous behaviors/operations are no longer an option. This research proposes to derive and prove the viability of a robust navigation architecture , which, as will be shown, is a key enabler for obtaining valid and reliable vehicle state estimates through correct interpretation of sensor measurements, even without the presence of GPS information. In a robust navigation approach, an autonomous vehicle navigates and estimates its states relative to a local coordinate system, which is based upon the available sensor information. When navigation with respect to that local coordinate system becomes difficult, the vehicle changes the frame of reference and continues navigation. In the background and out of the control loop, the vehicle can compute a globally-consistent map, localize within that map, and compute higher-level optimization-based behaviors. The benefits of such an architecture are: • The vehicle is natively able to utilize whatever sensor measurements are available to it. • The observability and consistency of the vehicle state can be maintained without the need for GPS position updates. • The approach is robust and not dependent on any one sensor. • The architecture should support the ability of multiple vehicles to contribute to the same mission and map. • The architecture can incorporate other types of information, such as known landmarks, or semantic constraints obtained from reading signs. Increasingly, the Air Force, and Department of Defense in general, are searching for methodologies to reduce their weapon system’s dependence on GPS. The purpose of the robust navigation approach is to allow a navigating autonomous /semi-autonomous vehicle to rely on other navigation approaches, thus enabling autonomous missions in adversarial environments where GPS information will be limited or completely unavailable. The initial work that will be described in the paper will derive the baseline relative navigation system for a fixed-wing UAV using visual odometry. The work will be accomplished in the coming months and is summarized by the following tasks. These task are presented in chronological order: • Derive vehicle dynamic and measurement models for use in a new indirect extended Kalman filter. • Validate the appropriateness of the models through the use of nonlinear observability analysis, which will be done using Lyapunov Stability theory. • Utilize a newly available filter implementation software tool package developed by the ANT Center, known as Scorpion, to implement the new filter and measurements equations. After the project, the framework and filter will be available as part of the tool set. • Simulate the performance of the algorithm with hardware-in-the-loop simulations, and if available, use real data sets to validate and tune the performance