An L
1
Adaptive Pitch Controller for Miniature Air
Vehicles
Randal W. Beard and Nathan Knoebel
Brigham Young University, Provo, Utah, 84602, USA
Chengyu Cao and Naira Hovakimyan
Virginia Polytechnic Institute and State University, Blacksburg, Virginia, 24061-0203, USA
Joshua Matthews
Raytheon Missile Systems, Tucson, Arizona, 85734, USA
One of the challenges in designing low level control loops for Micro Air Vehicles (MAVs)
is that the manufacturing process for airframes is not consistent enough to ensure uniform
aerodynamic properties. Therefore, there is a significant need for robust adaptive control
techniques that are computationally simple. Conventional Model Reference Adaptive Con-
trollers (MRAC) have proved to be very useful in a number of flight tests over the past
years. However, a major drawback of this control architecture is that during the transient
the control signal or the system output can exhibit large oscillations. This requires in-
tensive Monte-Carlo testing for all possible variations in all possible scenarios before each
flight test. This paper presents preliminary results for a novel adaptive control technique
that is both computationally simple, and has uniform bounded transient response. The
effectiveness of the proposed control scheme is demonstrated through simulation results
produced by a medium-fidelity hardware-in-the-loop simulator as well as flight test results
on a five foot wingspan unmanned air vehicle.
I. Introduction
The availability of small low power microprocessors and advances in solid state sensor technology have
enabled the development of small autopilots that have made it feasible to develop autonomous micro air
vehicles (MAVs).
1, 2
However, for MAVs to be economically viable, the cost of the electronics and the
airframe must be significantly less than conventional UAVs. There are currently a large variety of MAV
airframes, many of which are modified RC hobby aircraft. For most of these airframes, the aerodynamic
coefficients have not been suitably characterized and doing so is not economically feasible. The manufacturing
processes associated with MAV airframes are not precise enough to guarantee uniformity in the aerodynamic
coefficients. In addition, MAVs are designed to withstand frequent crashes, however the crashes often change
the aerodynamic properties of the airframe.
Therefore, the design of control systems for MAV cannot depend on accurate knowledge of the aero-
dynamic coefficients. An obvious solution is to use adaptive control techniques for the inner loops of the
autopilot. A variety of adaptive control techniques have been proposed for air vehicles including, neural
networks, least squares estimation, and Lyapunov based methods. The neural network approach typically
entails training a network off-line for model inversion, and then using an on-line adaptive network to com-
pensate for modeling errors.
3–7
Recursive least squares techniques identify the airframe parameters on-line
and use these parameters to adjust the controller. Such controllers have the ability to quickly converge on
the airframe parameters,
8
and therefore adapt quickly to failures. Other approaches to aircraft autopilot
design involving least squares estimation.
6, 9, 10
In Lyapunov based approaches, the parameter update law is
selected to ensure stability of the tracking error, but generally does not ensure parameter convergence.
11–14
Adaptive control for MAVs is still in its infancy. The Model Reference Adaptive Control (MRAC)
methodology was recently used to design roll and pitch attitude hold loops for MAVs in Ref. 15,16. Initial
flight tests demonstrated the feasibility of the method, however experimental implementation also highlighted
1 of 12
American Institute of Aeronautics and Astronautics
AIAA Guidance, Navigation, and Control Conference and Exhibit
21 - 24 August 2006, Keystone, Colorado
AIAA 2006-6777
Copyright © 2006 by the American Institute of Aeronautics and Astronautics, Inc. All rights reserved.