Implementing and Testing aNonlinear Model Predictive Tracking Controller forAerial Pursuit/Evasion Games on a Fixed Wing AircraftJ. Mikael Eklund, Jonathan Sprinkle and Shankar SastryAbstract— Flight test and simulation results are presentedfor a Nonlinear Model Predictive Tracking Controller (NMPC)used in pursuit and evasion maneuvers in three dimensionson a fixed wing Unmanned Aerial Vehicle (UAV) for thepurposes of pursuit/evasion games (PEGs) against a pilotedF-15 aircraft. These controllers are shown to be effective forboth asymmetric and symmetric PEGs. While the capability ofUAVs to perform autonomously has not yet been demonstrated,this is an important step to enable at least limited autonomyin such aircraft to operate with temporary loss of remotecontrol, or when confronted with an adversary or obstaclesfor which remote control is insufficient. Such capabilities havebeen under development in the Software Enabled Control(SEC) program and were recently tested in the CapstoneDemonstration of that program.I. INTRODUCTIONUnmanned Aerial Vehicle (UAV) have recently beenused with great success in gathering military intelligence[1] by providing a viable alternative to manned aircraftthrough their smaller size, reduced risk to life and limb,and reduced cost. These successes and challenges havestimulated research into UAV autonomy.UAVs have, however, exhibited very little autonomy to-date, and in the face of adversaries this lack of autonomyis a liability. While some defense against ground basedadversaries can be achieved through either low-level orvery high-level flight, success against an intelligent (i.e.manned) airborne adversary must rely on one of fourpossible dimensions in which to obtain an advantage: speed,maneuverability, munitions, and intelligence of control. Thisexperiment focuses on the last of those by improving theintelligence of the aircraft, which allows for current aircraftdesigns to be reused with software changes.Nonlinear model predictive control (NMPC) is a controltechnique that explicitly addresses nonlinear systems withconstraints on operation and performance. Previously theuse of NMPC has been shown to be effective for rotary-wing UAVs [2]. Aerial vehicles, with their nonlinear dynam-ics and input/state constraints to guarantee adherence to safeflight, are a proving ground for this technology. Although,the use of these control methods that run in real-time onfixed-wing UAVs has been in development [3], this has notbeen previously demonstrated in flight test. This is also theThis research was supported under DARPA’s IXO SEC program, undercontract number DARPA SEC F33615-98-C-3614.Department of Electrical Engineering and Computer Sciences,University of California, Berkeley. Berkeley, CA 94720, USA{eklund,sprinkle,sastry}@eecs.berkeley.eduFig. 1. The Global Hawk Medium Altitude Long Endurance UAV (photocourtesy of US Department of Defense).first application of NMPC to a symmetric PEG in whichthe evader aircraft is able to switch roles and become thepursuer.In this paper, the results of the final integration andtesting for the Capstone Demonstration of the SoftwareEnabled Control (SEC) program are presented for the fixedwing pursuit/evasion games (PEGs). A numerically efficientnonlinear model predictive tracking control (NMPTC) al-gorithm is used to encode the PEG between two fixed-wing adversaries. This follows on earlier work [3] andthe approach of [4]. The control problem is formulatedas a cost minimization problem in the presence of inputand state constraints. The minimization problem is solvedwith a gradient-descent method, which is computationallylight and fast [4]. The NMPTC controller uses an interfaceto an existing autopilot in order to influence the systembehavior. By formulating the cost function to include thestate information of the other aircraft, input saturation, stateconstraints and flight test boundary constraints, we showthe performance of the NMPTC as a one-step solution fortrajectory planning and control of UAVs competing in aPEG.This paper describes the experimental results of flighttests using an NMPTC controller that was designed toperform evasive maneuvers on a fixed-wing UAV whenconfronted by an airborne adversary of a priori type.Section II gives a brief description of the details of the fixed-wing aircraft used for flight and testing, and the expressionof the UAV’s dynamic and kinematic description. SectionIII describes the rules of the PEG. Section IV gives adescription of the encoding of the PEG requirements into2005 American Control ConferenceJune 8-10, 2005. Portland, OR, USA0-7803-9098-9/05/$25.00 ©2005 AACCWeC11.31509the controller for both pursuer and evader roles. Section Vgives the results of some games using the controller, andSection VI presents our conclusions and continuing work.II. DEVELOPMENT AND TEST PLATFORMSA. Aircraft detailsA Boeing Aircraft Company owned T-33 (originallymanufactured by the Lockheed Martin Company) two-seaterjet trainer was modified by Boeing for use in the live flighttesting in June 2004, and functioned as a UAV surrogateaircraft. The T-33 included a third party autopilot systemwhich did not include airspeed control of the aircraft. Thisaircraft will hereafter be referred to as the UAV. The UAVhad on board a safety pilot who could take control ofthe aircraft in the event of controller malfunction or poordecision making, and well as for controlling the airspeedof the aircraft based on indicator alerts and a displayedtarget airspeed given to the pilot to increase/decrease thrust.The route and trajectory of the UAV was controlled byCORBA-based experimental Technology Developer (TD)applications running on a laptop PC with a Linux operatingsystem and Boeing’s Open Control Platform (OCP) that wasinterfaced to the avionics of the aircraft. The TD applica-tions sent the control commands to the avionics pallet thattransformed them into autopilot maneuver commands. Thestate of the UAV, as well as the state of the other aircraft (anF-15 which exchanged state data with the UAV on a wirelesslink), was available via this avionics interface. The detailsof the avionics interface, the available state information, andthe input controls are given in the rest of this section.The NMPTC for PEGs described in this paper was oneof the TD applications developed and tested for this UAVas part of this SEC Capstone Demonstration program.B. Software in-the-loop simulation testbedIn order to facilitate