Rapid prototyping flight test environment for autonomous unmanned aerial vehicles

Transcription

1 Loughorough University Institutionl Repository Rpid prototyping flight test environment for utonomous unmnned eril vehicles This item ws sumitted to Loughorough University's Institutionl Repository y the/n uthor. Cittion: LIU, C.... et l, 211. Rpid prototyping flight test environment for utonomous unmnned eril vehicles. Interntionl Journl of Modelling, Identifiction nd Control, 12 (3), pp Additionl Informtion: This rticle ws pulished in the seril, Interntionl Journl of Modelling, Identifiction nd Control [ c Inderscience Enterprises]. The definitive version is ville t: Metdt Record: Version: Accepted for puliction Pulisher: c Inderscience Enterprises Plese cite the pulished version.

2 This item ws sumitted to Loughorough s Institutionl Repository ( y the uthor nd is mde ville under the following Cretive Commons Licence conditions. For the full text of this licence, plese go to:

3 Int. J. Modelling, Identifiction nd Control, Vol., o., 1 Rpid Prototyping Flight Test Environment for Autonomous Unmnned Aeril Vehicles Cunji Liu, Jonthn Clrke, Wen-Hu Chen nd John Andrews Deprtment of Aeronuticl nd Automotive Engineering Loughorough University Loughorough, UK {C.Liu5, J.H.A.Clrke, W.Chen, J.D.Andrews}@loro.c.uk Astrct: Test fcility is essentil for most of engineering reserch ctivities, from modeling nd identifiction to verifiction of lgorithms/methods nd finl demonstrtion. It is well known tht flight tests for erospce vehicles re expensive nd quite risky. To overcome this, this pper descries rpid prototyping pltform for utonomous unmnned eril vehicles (UAV) developed t Loughorough University, where numer of unmnned eril nd ground vehicles cn perform vrious flight nd other missions under computer control. Flexiility, mintinility nd low expenses re ssured y proper choice of vehicles, sensors nd system rchitecture. Among mny other technicl chllenges, precision nvigtion of the unmnned vehicles nd system integrtions of commercil-off-the-shelf components from different vendors with different opertionl environments re discussed in detil. Mtl/Simulink sed softwre development environment provides semless rpid prototyping pltform from concept nd theoretic developments to numericl simultion nd finlly flight tests. Finlly, two scenrios performed y this test fcility re presented to illustrte its cpility. Keywords: UAV; utonomous; re-ltime; flight control; pltform. Biogrphy notes: Cunji Liu received BEng in Detection, guidnce nd control technology (25), MEng in Guidnce, nvigtion nd control (28) from Beijing University of Aeronutics nd Astronutics, Beijing, Chin. He is currently PhD student in the Deprtment of Aeronuticl nd Automotive Engineering t Loughorough University, UK. Jonthn Clrke Received MEng in Aeronuticl Engineering (28) from Loughorough University. He is currently PhD student in the Deprtment of Aeronuticl nd Automotive Engineering t Loughorough University, UK. His current reserch interests include prmetric modelling of shllow cumulous convection, nd utonomous soring nd pth plnning. Wen-Hu Chen holds Senior Lectureship in flight control systems with the Deprtment of Aeronuticl nd Automotive Engineering, Loughorough University, Leicestershire, U.K. Before this, he ws Lecturer with the Deprtment of Automtic Control, njing University of Aeronutics nd Astronutics, njing, Chin from 1991 to 1996, nd he held reserch position nd then Lectureship in control engineering with the Centre for Systems nd Control, University of Glsgow, Glsgow, U.K from 1997 to 2. He hs pulished one ook nd more thn 1 ppers in journls nd conferences. His reserch interests re the development of dvnced system nd control strtegies nd their pplictions in erospce engineering. John Andrews is Professor of Systems Reliility in the Deprtment of Aeronuticl nd Automotive Engineering t Loughorough University, UK. The prime focus of his Copyright 28 Inderscience Enterprises Ltd.

4 2 C. Liu, J. Clrk, WH. Chen, J. Andrews reserch hs een on methods for predicting system reliility in terms of the component filure proilities nd representtion of the system structure. Much of this work hs concentrted on the Fult Tree technique nd the use of the Binry Decision Digrms (BDDs) s n efficient nd ccurte solution method. He is the uthor of round 2 reserch ppers on this topic nd is joint uthor, long with Bo Moss, of text ook, Reliility nd Risk Assessment, now in its second edition, pulished y ASME. John is Founding Editor of the Journl of Risk nd Reliility (prt O of the IMechE Proceedings). He is lso memer of the Editoril Bords for Reliility Engineering nd System Sfety, nd Qulity nd Reliility Engineering Interntionl 1 Introduction Recently incresing reserch hs een devoted to the field of Unmnned Aeril Vehicles (UAVs) since it is widely elieved tht numerous civil nd militry pplictions cn e found for UAVs (Deprtment of defence, 27). To develop nd improve intelligence nd utonomy of UAVs, dvnced methodologies rnged from individul flight control, multi-vehicle control nd coordintion to mission plnning nd decision mking hve eing developed. These lgorithms need to e evluted nd verified in order to ssess their prcticl performnce, nd pve the wy for inserting them into rel engineering prctice. For this purpose unique indoor rpid prototyping pltform hs een developed t Loughorough University. This rticle descries the detils of the pltform nd presents some test results to illustrte its cpilities. It is well known tht flight tests re very expensive, impose high risk for personnel nd ssets, nd requires lrge irfield nd hevy logistic supports. Due to these resons, most of the reserch nd development works on erospce vehicles such s ircrft, missiles, nd rotorcrfts re still evluted y numericl simultion. This hs een identified s one of the min ostcles for trnsferring dvnced control concepts nd methods into rel engineering prctice. On the other side, UAV hs een one of the most ctive reserch topics driven y numerous militry nd civil interests. Although there is significnt progress in the reserch of concepts, there is lck of test fcilities to verify these new methodologies. It is impertive to hve proper test fcility to fcilitte these reserch ctivities nd de-risk the new reserch ides generted from these ctivities. To ddress this issue, numer of ttempts hve een mde recently, including vrious hrdwre-inthe-loop simultion nd flight testing fcilities. At Georgi Institute of Technology, n open system UAV tested referred to s RTMx ws developed to investigte flight control lgorithms (Johnson et l., 24). Reserchers in University of Cliforni t Berkeley use pltform comprised of fleet of commercilly ville rotry-wing nd fixed-wing UAVs to study pplictions such s utonomous explortion in unknown urn res (Shim et l., 25). In the Aerospce Controls Lortory t the Msschusetts Institute of Technology (MIT), n outdoor tested consisting of fleet of eight fixed-wing utonomous UAVs provides pltform for evluting coordintion nd control lgorithms (King et l., 24). In generl, these outdoor pltforms provide most relistic flight tests, however they suffer numer of drwcks. Firstly, they re expensive nd hve limittions on how quickly they cn perform flight tests due to the constrints on ccessing lrge irfield. Secondly, most outdoor unmnned eril vehicles cn only e flown in good wether conditions to void risks. Thirdly, UAVs typiclly require lrge logistic support tem, which mkes testing logisticlly difficult nd expensive. In contrst, indoor testeds my provide much more flexile, ccessile nd cheper fcility for UAVs nd for generl flight control reserch. The min constrints for indoor testeds re confined spce nd strict requirements on vionic systems. To this end, MIT s RAVE tesed is the most promising indoor tested, where n environment with numer of qud-rotor ircrft hs een developed to investigte long durtion missions nd helth mngement (How et l., 28). Although most of the hrdwre components re commercil-off-shelf prts, the softwre environment Open Control Pltform ws initilly developed nd provided y The Boeing Compny. An indoor rpid prototyping pltform is descried in this pper to perform vrious flight tests nd other multivehicle coordintion. It cn verify control level nd mission level lgorithms into the rel world in semless wy nd speed up the process from theory to prctice. The min fetures of this tested re: Flexiility nd verstile. Almost ll the commercil model vehicles nd rotorcrfts cn e operted in this environment. Low costs. It is designed sed on the commercil-offthe-shelf concept. User friendly. Without eing n expert in coding nd electronics, reserchers using this pltform cn focus on theoretic methods nd lgorithms. Rpid prototyping. This llows reserchers to strt from lgorithm development to numericl simultions to finl rel-time flight tests on the corresponding unmnned vehicles. 2 Pltform rchitecture nd components 2.1 Design chllenges nd philosophy One ojective of the rpid prototyping pltform is to enle reserchers to test vriety of lgorithms pplicle to UAV coordintion nd control in nerly rel-world scenrios. So vehicles with good hndling qulity nd

5 Rpid Prototyping Flight Test Environment for Autonomous Unmnned Aeril Vehicles 3 mnoeuvrility re needed in this pltform. Another ojective is to simplify the opertion on demonstrtions nd llow n individul to crry out the entire process from lgorithm design to evlution. Tht lso mens tht components tht require less modifiction, hve good reliility nd re esily mintined. To meet these demnds, the pltform dopted proper commercil-off-the-shelf equipments nd comines them effectively. The rchitecture of the pltform follows simple hierrchicl design. The key constrint of indoor test fcility for flight tests is tht the opertion spce is very limited. So only very smll unmnned vehicles cn e used to perform vrious relistic flight tests, rther thn just tking off nd lnding. This implies very little pylod or no pylod cn e put on these smll eril vehicles. The core technique in the pltform is n oject trcking system. Tht llows ground sttion computer to perceive the position nd ttitude of vehicles in the test re, insted of mounting n onord computer nd sensor suite on vehicles in conventionl wy. The lowcost off-the-shelf rdio controlled (R/C) vehicles cn e used in the pltform without modifictions since there is no significnt pylod requirement on the eril vehicles. In ddition, ll these seprte components were finlly integrted into the Mtl/Simulink environment tht is widely used in cdemi nd industry for reserch nd development. Both high level utonomous lgorithms nd low level control lgorithms re uilt in the ground sttion, ut control lgorithms re modulrized to custom to different vehicles. Modulriztion llows esy ddition or removl of different types of vehicles, s needed for different scenrios. Ech low level control model hs the cpility to ccess hrdwre directly to enle their functions. The structure of the pltform is shown in Fig. 1. Figure 1 Pltform structure vehicles for prticulr ttention. It is fixed pitch electric helicopter with reltively low rotor tip speed, which mens tht there is less energy in the min rotor system mking the helicopter considerly sfe to operte in n indoor environment. Its plstic components cn lso e esily replced fter crsh. Moreover, the Hummingird helicopter is not only chep nd commercilly ville form most R/C model shops, ut is the one of est hndling 3 clss indoor helicopters. All these resons mke the Hummingird good choice s n utonomous helicopter. In ddition, the Hummingird helicopter hs een modified to use LiPo tteries to increse the flight times. The ltertion of the ttery type mens lrge reduction in weight nd ner douling of the flight times. Figure 2 Hunmmingird Helicopter In ddition to the hummingird helicopters, there re numer of other helicopters nd quod-rotors used s eril test vehicles. The ground vehicles dopted re Tmiy TT1 crs, which is type of R/C electric model cr. These eril nd ground vehicles enle the user to construct vrious scenrios such s formtion, surveillnce, trcking, nd so on. It shll e highlighted tht the dynmics nd control mechnisms of these helicopters nd ground vehicles re very much the sme to the norml ones except the scle or the chnge of certin coefficients. 2.3 Vicon motion system The Vicon motion cpture system provides powerful trcking fcility suitle for the indoor environment, which uses MX cmers to sense the lightweight reflective lls or elts in the operting re (Vicon, 28). Therefore, y ttching some reflective mrkers to vehicle, the vehicle cn e detected y Vicon cmers. The mrker position informtion is then trnsmitted vi Ethernet using TCP/IP to computer where Vicon exus softwre clcultes the position nd orienttion of the vehicle. The MX cmer nd exus re shown in Fig. 3. The rpid prototyping pltform minly consists of three components, which re smll-scle eril or ground vehicles, the Vicon motion system nd ground sttion. 2.2 Aeril/ground vehicles Smll-scle vehicles ply n importnt role in indoor tests for emulting the ehviour of rel UAVs nd setting vrious scenrios. These vehicles in the pltform, especilly eril vehicles, need to e low cost, low risk ut e highly flexile nd hve good mintinility. As result rdio controlled helicopter, clled Hummingird (Fig. 2), ws chosen s one of the test Figure 3 Vicon cmer nd exus softwre Currently 5 MX nd 3 T1 cmers re used to cover 5m y 4.5m y 2m testing volume. The sttic ccurcy ws

6 ssessed y mesuring the position of helicopter sitting on the floor. The result in Fig. 4 shows tht the 2 hours drift is less thn.25 mm. Due to the principle of the motion cpture system, this gives fir indiction of the position nd ttitude ccurcy of the motion-cpture ystem during flight opertions. The Vicon system cn cpture n oject motion with refresh rte up to 1Hz nd cn still trck the vehicle even if one or two mrkers on vehicle re missing. Therey the Vicon system cn e regrded s high ndwidth nd roust nvigtion system in this pltform. It cn e considered s n indoor replcement of Glol Positioning Systems (GPS) ut providing not only the position ut lso ttitude informtion.. Distnce [mm] Figure 4 Drift over 2 hours Ground sttion Time [Minuits] 2 hours drift of Vicon system Ground sttion consists of severl personl computers (PCs) with Intel Core2 66 CPU t 2.4GHz nd some ccessories. The ground sttion cts s the rin of the pltform, ecuse the control nd commnds re sent from the ground sttion fter tll the clcultions re performed there. To control vehicles in the test re, these computers run Vicon exus nd Mtl pplictions which provides the position nd ttitude informtion of vehicles nd clcultes corresponding control commnds sed on utonomous lgorithms respectively. In terms of high-level tsks, ground sttion lso mnges tsks such s mission plnning nd trjectory design. In order to send control signls to the vehicles, the ground sttion is equipped with JR9X2 computer trnsmitters. The ridge etween the computers nd trnsmitters is n dpter, whose one end connects to the computers vi USB port nd the other end to the R/C trnsmitters through triner port interfce s shown in Fig. 5. The proper selection of components is the key to success in uilding n dvnced UAV test pltform. 3 System integrtion sed on mtl / simulink This section descries the detils of integrting oth hrdwre nd softwre to construct rpid prototyping environment sed on Mtl/Simulink. Although the commercil-off-the-shelf components provide necessry functionlities for the pltform, considerle systemintegrtion effort is required to link ll these components together within single environment despite they my hve softwre drivers/pckges developed y different source codes. Usully this environment should e governed y rel-time opertion system or similr softwre which enles ll components to communicte with ech other in rel-time during tests. 3.1 Initil fesiility nlysis Since multiple susystems re involved in the pltform, it is essentil to synchronize the execution mong these susystems in order to gurntee dt comptiility, prticulrly when relistic helicopter or ground vehicle needs to e controlled. However, there exists chllenge tht ech hrdwre product hs its own softwre or driver nd is operted independently. Vicon motion system processes dt cptured y MX cmers using softwre, nmely exus, which contins rel-time engine providing processed position nd ttitude informtion of ojects. This engine cn only e ccessed through Vicon rel-time Appliction Progrmming Interfce (API) written in C lnguge. Moreover, the dpter connected to the computer through USB port is driven y C++ style API under Windows operting system (OS). Fortuntely the process in which the rdio trnsmitter trnsmits commnd signls form PCTx dpter to helicopter is strightforwrd nd needs no modifiction. To chieve our purpose, Mtl/Simulink is used to uild the softwre environment to mnge ll the hrdwre. Mtl/Simulink is very powerful nd convenient tool for control system design nd simultion, which lso provides numer of communiction mens nd mechnism of integrting with C/C++ lnguge. On the other hnd, UAV utonomous lgorithms, t the core of the pltform re usully developed nd implemented y utilizing Mtl/Simulink. This mkes it very promising cndidte for the semless trnsition from design to numericl simultion nd rel-time vlidtion in single softwre environment. To overcome the prolem tht the common Mtl/Simulink progrms re executed in computer time rther thn rel-time, two different rel-time implementtion environments sed on Mtl/Simulink were tested y utilizing different techniques. Figure 5 Trnsmitter nd dpter

7 Rpid Prototyping Flight Test Environment for Autonomous Unmnned Aeril Vehicles Rel-time environment using rel-time lockset An integrted implementtion environment requires the cpilities of communiction nd execution in rel-time. Simulink provides powerful mechnism for extending its cpility, nmely S-function, which is computer lnguge description of Simulink lock. S-function cn e written in C/C++, which cn ccess to the Vicon API receiving vehicle stte dt nd cn drive the dpter sending commnd signls. Thus, two dedicted S-function locks were developed to communicte with the Vicon system s well s the dpter within Mtl/Simulink environment. Despite using C/C++ during the development progress, the completed locks cn e treted s common Simulink locks nd re esy to use. After solving communiction prolems, rel-time lock set ws dded into the Simulink environment to ensure the implementtion in rel-time (Leonrdo, 28). This lock set minly holds the execution of the Simulink simultion to the rel-time. If the cycle time is lower thn the simultion step, this lock set wits for the time needed to fill the simultion step, leving the remining CPU time to ll the other Windows Processes or just idle. This concept is very simple ut effective. During the rel-time tests, Vicon exus nd Simulink run in the sme PC tht connects Vicon MX cmers through Ethernet nd the dpter thought USB port respectively. In this mnner, the time dely of dt trnsfer cn e minimized. The ltency of the clculted Vicon dt due to the network is less thn 1ms, while the ltency of sending out control signls is out 5ms on verge due to the property of the USB port. Currently the environment is running t the smpling intervl of 1ms, nd typicl tsk execution time (TET) with norml set of control nd commnd lgorithms is given in Fig. 6. The lower line represents the dely in the execution of ech time intervl, while the upper line represents the remining time for which Simulink wits for the next step (nd leve the CPU to remining Windows pplictions). Figure 6 TET of typicl control lgorithm The softwre environment sed on the Simulink cn mnge the dt trnsfer utonomously in the ckground. Therefore one cn use this pltform s if it is norml Simulink environment. A detiled structure of this environment is shown in Fig. 7. Figure 7 Structure of rel-time lock set environment The dvntges of this environment re tht it provides mny powerful tooloxes nd other useful uilt-in resources in Simulink, nd it is lso very convenient to oserve nd record signls during the flight tests. Furthermore, it ccelertes the development significntly, ecuse there is no ostcle etween lgorithm development nd rpid prototyping. One cn implement lgorithms into this pltform for experiments directly s long s the numericl simultion completed in the Simulink. evertheless, there re some drwcks, which might hve influence on flight tests for some scenrios in the future. For more complicted mission, due to the hevy computtion urden, it might e difficult to complete the clcultion within the smpling intervl when the RT Block set is used. Furthermore, from the operting system point of view, Mtl/Simulink is running on Windows OS, which is not true rel-time operting system (RTOS). Tht mens tht other pplictions with higher priority in Windows system my interrupt rel-time experiments nd possily cuse the loss of helicopters. To void these negtive spects, nother rel-time environment sed on xpc Trget ws then developed. 3.3 xpc Trget environment xpc Trget is specil product in Mtl for prototyping, testing, nd deploying rel-time systems using stndrd PC hrdwre. It is n environment tht uses trget PC, which seprtes from host PC, for running rel-time pplictions. Theses pplictions re creted from Simulink progrm on the host PC nd downloded into the trget PC through Ethernet or seril connection. xpc Trget cn significntly enhnce the reliility nd hve the cpility of deling with more complicted lgorithms. The structure of xpc Trget environment is different from the previous one. Since the rel-time execution is chieved essentilly, the synchronising communiction is the remining issue tht is of concerned in this structure. xpc Trget executes its pplictions on rel-time kernel, where Vicon exus is not comptile nd USB port is not supported. Therefore the trget PC hs to communicte with nother server PC tht cn provide the vehicle s sttes clculted y Vicon nd send commnd signls clculted y the trget PC to trnsmitters. There re two sic communiction methods uilt in xpc Trget. One is RS232 seril port trnsport, while nother is User Dtgrm Protocol (UDP) technology. The ltter one is chosen in our cse, ecuse of its high ndwidth nd the ility of tlking to multiple nodes in the network. Although

8 UDP protocol elimintes error check nd recovery, it ensures tht rel-time pplictions hve mximum chnce of succeeding in rel-time execution y only using the most recent dt. On the other hnd, one should locte the trget PC, host PC nd server PC within locl re network (LA) to minimize the network ltency. In ddition, C/C++ server progrm ws developed running on the server PC. This progrm tkes chrge of the dt trnsmission etween trget PC nd server PC, where Vicon dt re converted to UDP pckets to send out, menwhile received UDP pckets re decoded into control signls to drive the trnsmitter. The synchroniztion of dt trnsfer is implicitly delt with in mnner tht the min ppliction on the trget PC clls ech communiction port t fixed intervl during test, wheres the server progrm receive nd send pckets pssively. The entire structure of this environment is shown in Fig. 8. Figure 8 Structure of xpc Trget environment When doing the rel-time implementtion using this environment, the user needs to configure the network mong these PCs first, nd then compile Simulink progrms into executle rel-time pplictions nd downlod to the trget PC. This work is implemented in the Mtl environment. During flight tests, vehicle sttes cn e visulized y the Vicon exus on the server PC or cn e displyed in numericl or curve form on the monitor of the trget PC. After flight test ll the dt cn e logged ck to the host PC for recording or nlyzing. The merits of xpc trget environment re tht it provides considerly more relile environment for implementing vrious lgorithms nd hs the potentil of expnsion to meet the rel time requirement for sophisticted lgorithms. Although the opertion of this environment is not s esy s the rel-time lock set environment, it is very suitle to demonstrte reltively mture lgorithms into complicted scenrio such s multiple vehicle coordintion nd long durtion mission mngement. 4 Possile usege of the flight test environment The flight test fcility descried ove is very verstile nd flexile, nd provides support for mny ctivities s outlined elow. 4.1 Reserch ctivities It cn provide support for the following reserch ctivities: System identifiction nd modelling Modelling is lwys the first step in developing control nd other strtegies (Shrm, 29). Vrious tests of the helicopters nd ground vehicles cn e performed under humn remote control. All the control commnds such throttle nd the response of the vehicles cptured y the opticl trcking system cn e recorded. This provides n idel environment for identifiction nd modelling. This function is very much similr to wind tunnels for fixed-wing ircrft Flight control Helicopters hve very complicted dynmics, with strong nonlinerities nd coupling etween different chnnels. To some extend, control of smll scle helicopters is even more chllenging thn tht of conventionl helicopters since they re more susceptile to ground effects nd the chnge of structure nd propulsion. Vrious control lgorithms cn e developed using dvnced control methodologies such s nonliner control nd roust control nd then evluted in this flight test environment. The control clcultions re performed in Mtl/Simulink in norml PCs, which not only eses the implementtion ut lso provides enough computing power required for complicted lgorithms such s model predictive control where one-line optimiztion is required Avionic systems vigtion systems re very importnt prt of the onord vionic systems, nd provide essentil informtion for ircrft control nd positioning (Pnzieri et l., 28; Frvolini et l., 28). The opticl trcking system cn e used s reference system to ssess the performnce of vrious new nvigtion systems. For exmple, one reserch topic is to investigte the integrtion of low cost inertil mesurement sensors with computer vision. Together with inertil sensors, smll cmer cn e instlled on the helicopter to perform vrious flight tests to investigte the performnce of these new concepts nd lgorithms. It cn provide support for similr work on vehicle nvigtion systems Autonomous lgorithms on pth nd mission plnning One of the most min motivtions for developing this flight test environment is to support reserch in utonomous lgorithms such s mission plnning nd pth plnning. In these high level lgorithms, the UAVs re treted s mss point, so the lgorithms re lrgely independent of the pltforms. To this end, this test fcility provides n environment to verify nd de-risk reserch work on UAV utonomy for vrious eril vehicles including fixed wing ircrft. 4.2 Teching This flight test environment lso provides support for teching ctivities. Two modules Flight control Systems nd Avionic Systems directly enefit from it y setting the

9 Rpid Prototyping Flight Test Environment for Autonomous Unmnned Aeril Vehicles 7 coursework nd hving experimentl tests in this environment. It hs lso een used to support vrious finl yer projects nd other group design projects. 4.3 Demonstrtion It is importnt to promote university reserch ctivities to the pulic, the stkeholders nd vrious visitors. The test fcility descried in this pper is now one of the most populr ls for touring round the Deprtment. It is now well received from student pplicnts, industril prtners nd other visitors. This not only promotes the existing outcomes ut lso helps to ttrct potentil funds nd collortions for expnding current reserch ctivities. 5 Rel-time control tests Before high level lgorithms re pplied to the pltform, it is necessry to complete vehicle stiliztion nd control functions. This is ecuse effective nd relile control lws re essentil for mission-level tsks. In this section, PID controller for Hummingird helicopter is demonstrted using rel-time lock set environment. Then simple control test using two crs where they trck two circle trjectories simultneously is presented to illustrte the rpid prototyping ility of the xpc environment for multivehicle situtions. 5.1 Helicopter flight control In order to study the chrcteriztion of the helicopter nd design the controller, linerized model for hovering sttus ws creted, which cn e written in the stte-spce form (Mettler et l., 1999): x & = Ax + Bu (1) with the stte vector x nd the control u given y: x = u = [ u v p q φ θ w r r f ] [ δ δ δ δ ] T lt lon The stte vector includes 11 vriles: liner velocities in x, y nd z directions tht re u, v nd w, respectively; ttitude vriles: roll φ nd pitch θ ; ngulr rtes for roll, pitch nd yw: p, q nd r; rotor longitudinl nd lterl flpping ngle: nd ; yw rte gyro feedck r. In ddition, the four control inputs re longitudinl nd lterl cyclic δ lt nd δ lon, the pedl δ ped nd the throttle δ thl, respectively. The stte mtrices re lso shown elow: ped thl f T X u Lu M u A = B = Alt Blt Y L M A B v v lon lon v 1 1 p ped 1 1 g Z thl thl g X L M 1 τ B Z f Y L M A 1 τ Z f Z w w Z K r r r ped K rf where τ is the rotor time constnt, g is the ccelertion of f grvity, the other prmeters in mtrix A re stility derivtives nd prmeters in mtrix B re control derivtives. The model structure is developed through first principles, nd the vlues of prmeters re found y the system identifiction process (Shim et l., 2). Helicopters s descried y eqution (1) re difficult to control due to their inherent instility nd coupling effects. The outer loop of trnsltionl velocity u nd v re not only decided y the ttitude of the fuselge, ut influence y the rotor dynmics, which cn e reflected y the terms of X nd X. Moreover, the inner loop ttitude vriles re coupled with ech other through terms of L nd M. Although the rel helicopter shows coupling effects mong different chnnels, in the hovering sttus, the system cn e considered s four seprte susystems tht re roll, pitch, yw nd heve chnnels, with the coupling treted s form of disturnce. Therefore, the trnsfer function for this multiple SISO system cn e derived s following: φ θ ψ Z T ( s) = ( s) ( s) ( s) ( s) (2) δ lt δ lon δ ped δ thl Ech chnnel cn e controlled y conventionl PID controller (Cstillo et l., 25). The time-domin eqution of it is: de ( ) ( ) ( ) ( t) u t = K Pe t + K I e t dt + K D (3) dt e( t) = r( t) y( t) where e ( t) is the error etween the required nd the ctul signls. K P, K I nd K D re the proportionl, integrl nd derivtive gins, respectively. The overll flight control dopts cscded structure with two loops. The inner-loop controller is composed of four SISO controllers s descried ove to mke sure the helicopter following the required ttitude profiles. The T

10 outer-loop controller is used for guidnce purpose generting the ttitude commnds sed on the position of the helicopter nd the trjectory to trck. The controller structure is shown in Fig 9. Figure 9 Flight control structure for helicopters The proper gins for ech PID controller tht cn stilise the rottionl nd trnsltionl dynmics of the helicopter re needed. First, sed on the trnsfer function of ech chnnel in eqution (2), the rough rnges of the gin vlues were determined y using Mtl model-sed control design tools. Then these gins were empiriclly djusted during flight tests to get stisfctory performnce. of the helicopter remins inside the 2-cm ox during the hovering except the tkeoff stge. In the trcking test, the Hummingird ws controlled to follow the trjectory consisted of 5 wypoints. The result in Fig. 11 shows tht the helicopter cn trck pth t low speeds with mximum position error less thn 2 cm. 5.2 Multi-vehicle control To illustrte tht the xpc environment nd the ility to demonstrte coordintion lgorithms, two crs were controlled to trck two round trjectories with different dimeters. The configurtion of rer-wheel driving vehicle, s the rdio controlled model cr used in our l, is shown in Fig.12. Figure 12 Vehicle configurtion The stte vriles of the model re [ x y θ ] T, where ( x, y) re the coordintes of the centre point of the rer xle. θ is the heding ngle of the cr ody with respect to the x xis. Angle φ in Fig. 12 is the steering ngle of the front wheel with respect to the vehicle s longitudinl xis, which is control input. Another prmeter of the model is the distnce etween the front xle nd the rer xle, which is l. The kinemticl reltionship cn e descried using the following mthemticl model: Figure 1 Hovering result of the helicopter Figure 11 Trcking performnce of the helicopter The helicopter hovering test result nd simple trjectory trcking result re given inn Fig. 1 nd 11. The hovering result demonstrtes tht the helicopter cn hold it position during flight. In this test, the Hummingird took off, hovered t.5m ove the origin nd then lnding utonomously. The result in Fig.1 shows tht the position x& = v cosθ y& = v sinθ & tnϕ θ = v l where, control input u = [ ϕ v] T is the steering ngle nd line velocity. otice tht there is nonholonomic constrint relted to cr-like vehicle, which refers to the constrint of rolling without slipping etween the rer wheels nd ground, nd tht cn e represented s follows: ( θ + ϕ ) y& cos( θ + ϕ ) & θ cosθ l x& sin = (5) For the demonstrtion purpose the speed of ech cr is limited to constnt vlue, wheres the steering ngel is controlled y PID controller to follow circle. As shown in Fig. 13, lthough the strt points were not t the expected circles, the controller cn mke the crs following the trjectories. (4)

11 Rpid Prototyping Flight Test Environment for Autonomous Unmnned Aeril Vehicles 9 Figure 13 Two crs trcking 6 Conclusion In this pper, the hrdwre nd softwre structure of rpid prototyping flight test environment for utonomous UAVs is discussed in detil. This pltform provides convenient nd effective fcility for evluting UAVs nd generl flight control reserch ctivities using rel vehicles. This is very verstile tested nd vrious scenrios including single eril vehicle, multi-vehicles, mixed of eril nd ground vehicles cn e tested. As outlined in Section 4, it cn provides support for vriety of reserch, teching nd demonstrtion ctivities. The doption of the widely used Mtl/Simulink environment enles reserchers to test new reserch outputs semlessly on rel vehicles. This multifunctionl, low cost nd flexile indoor flight tested lso enles one resercher to mnge/coordinte missions with multi-vehicles, which significntly reduces mnpower nd logistic supports required for this kind of tsks. Another feture of this pltform is tht model helicopters re dopted s eril vehicles. Due to the nture of helicopter dynmics such s hovering, verticl tkeoff/lnding nd low speed cruise, this llows relistic nd complicted missions to e simulted in confined spce. Vrious scenrios such s helicopter hovering nd trcking, nd two ground vehicles coordintion nd helicopter trcking moving ground vehicle hve een successfully implemented on this pltform to illustrte its cpilities. How, J. P., Bethke, B., Frnk, A., Dle, D. nd Vin, J. (28) "Rel-time indoor utonomous vehicle test environment," IEEE Control Systems Mgzine, vol. 28, pp Johnson, E.., Schrge, D. P., Prsd, J. V. R. nd Vchtsevnos, G. J. (24), "UAV flight test progrms t Georgi Tech." Proceedings of the AIAA Unmnned Unlimited Technicl Conference, Workshop, nd Exhiit. King, E., Kuwt, Y., Alighnri, M., Bertuccelli, L., nd How, J. P. (24) "Coordintion nd control experiments on multivehicle tested," Proceedings of the Americn Control Conference. vol. 6, pp Leonrdo, D. (28) "RT Blockest for Simulink," Mettler, B., Tischler, MB. nd Knde, T. (1999) "System identifiction of smll-size unmnned helicopter dynmics," Annul Forum Proceedings of Americn Helicopter Society Pnzieri, S., Pscucci, F. nd Setol, R. (28) Simultneous loclistion nd mpping of moile root vi interlced extended Klmn filter Interntionl Journl of Modelling, Identifiction nd Control. Vol. 4, o.1 pp Shrm, S. (29) Adpttion of modelling nd identifiction: n eronuticl engineering perspective. Interntionl Journl of Modelling, Identifiction nd Control. Vol. 6, o.2 pp Shim, D. H., Chung, H., Kim, H. J. nd Sstry, S. (25) "Autonomous Explortion in Unknown Urn Environments for Unmnned Aeril Vehicles," AIAA G&C Conference, Sn Frncisco, pp Shim, D. H., Kim, H. J., nd Sstry, S. (2) "Control system design for rotorcrft-sed unmnned eril vehicles using time-domin system identifiction," Proceedings of the 2 IEEE Interntionl Conference on Control Applictions. Vicon, (28) "Vicon Motion Cpture Systems," Xi, W. & Brs, J.S. (27), "MPC sed motion control of crlike vehicle swrms", Mediterrnen Conference on Control & Automtion. References Cstillo, C. L., Alvis, W., Cstillo-Effen, M., Moreno, W. nd Vlvnis, K. (25) "Smll scle helicopter nlysis nd controller design for non-ggressive flights," Systems, Mn nd Cyernetics, IEEE Interntionl Conference on, vol. 4, pp Deprtment of Defence (25), USA. "Unmnned Aircrft Systems Rodmp: " Office of the Secretry of Defence. Frvolini, M. L., Cmp, G. nd politno, M. R. (28) Modelling nd performnce nlysis of mchine vision-sed semi-utonomous eril refuelling. Interntionl Journl of Modelling, Identifiction nd Control Vol. 3, o.4 pp