Implementation of a New PC based Controller for a PUMA Robot based on COEM

Transcription

1 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ Implementaton of a ew PC based Controller for a PUMA Robot based on COEM M. Farooq M, Member IAEG. Dao Bo Wang Abstract The Controller Output Error Method (COEM) s ntroduced and appled to the desgn of an adaptve fuzzy controller for a PUMA robot. The method employs a gradent decent algorthm to mnmze a cost functon whch s based on the error at the controller output. The cost functon s mnmzed by adaptng some or all of the parameters of the fuzzy controller. The paper also descrbes the replacement of a controller for a PUMA 52 robot wth a newly desgned PC based (open archtecture) controller employng COEM. The orgnal structure of the PUMA robot s retaned. The hardware of the new controller ncludes n-house desgned parts as: PWM amplfers; dgtal and analog controllers; I/O cards; sgnal condtoner cards; and 6 bts A/D and D/A boards. An Intel Pentum IV ndustral computer s used as the central controller. The control software has been mplemented usng VC++ programmng language. The traectory trackng performance of all the sx onts was tested at varyng veloctes. The expermental results showed that t s feasble to mplement the suggested open archtecture platform for PUMA 5 Seres robots through software routnes runnng on a PC. Index Terms PUMA robot, Adaptve fuzzy control, Pulse wdth modulaton (PWM) amplfer, Graphcal user Interface (GUI), Controller output error method (COEM). I. ITRODUCTIO Robots form an essental part of mechatroncs and Computer Integrated Manufacturng (CIM) systems. Robots are generally controlled by dedcated controllers. As upgrades become costly and nterfacng becomes complex due to hardware and software conflcts, the flexblty of the robotc manpulators s reduced. Dedcated hardware and propretary software whch normally allows only hgh level programmng by the users are costly and dffcult to understand. Snce the early years of 98 s many proects have been carred out to develop an open archtecture controller such as GC[], GISC[2], ROBLIE[3] and so on whch try to solve the problem of realzaton of an open archtecture controller, and several prototype systems have been developed. However, they are not wdely accepted due to the overly restrctve defntons and specal standards. Manuscrpt receved January 7, 28. Ths work was supported n part by the Smulaton & Machne Control Lab, College of Automaton Engneerng, anng Unversty of Aeronautcs & Astronautcs. M. Farooq s wth the Smulaton & Machne Control Lab, College of Automaton Engneerng, anng Unversty of Aeronautcs & Astronautcs, 26, PR Chna (phone: ; fax: ; e-mal: muhfarooq@nuaa.edu.cn) Dao-Bo Wang s wth the Smulaton & Machne Control Lab, College of Automaton Engneerng, anng Unversty of Aeronautcs & Astronautcs, 26, PR Chna (e-mal: dbwang@nuaa.edu.cn).. The Unmate PUMA 5 seres Robots manly uses DEC LSI processor runnng VAL robot control software[4]. Methods of bypassng VAL are dscussed n lterature, ncludng Unmaton techncal reports[5]. However, most of these procedures have been confned to replacng the LSI wth another DEC computer, leavng perpheral hardware ntact. A well- refned open structure archtecture for ndustral robot s dscussed n [6]. However; t s manly based on Common Obect Request Broker Archtecture (COBRA), leavng scope to smplfy the hardware and software work. A hardware retroft for Puma 56 robot s dscussed n [7] but stll t reles on specal-purpose TRC4 cards nstalled on the backplane of Mark II controller. The shft towards the personal computer open archtecture robot controller and the mpact of usng these newer controllers for system ntegraton s dscussed n [8]. In fact; t s far more cost effectve to develop new hardware usng less specfc nterfaces. An mproved PC-based desgn for Puma robot was presented n [9] but ths hardware confguraton purely depends on n-house bult desgns. In our paper a flexble, modular hardware s developed for the puma robot, ncorporatng a personal computer, n-house as well as specalzed hardware. Some techncal problems n the prevous desgn for velocty test profle of onts, 2 and 4 have also been addressed. The onts poston trackng error at hgh veloctes s also mnmzed n our desgn. In ths paper, an adaptve fuzzy control desgn s developed for robot traectory trackng control. The controller output error method (COEM) s ntroduced n proposed desgn, whch can be used for the on-lne tunng of parameters of a fuzzy controller. Ths method can be used wth any fuzzy controller desgn; the only requrement s that controller must be stable before the commencement of on-lne tunng. Thus, at frst, any fuzzy rule-based model and any nference mechansm can be employed to parameterze and ntalze controller of the system, and then COEM s appled subsequently to tune parameters of fuzzy controller for purpose of achevng better performance Although the method has already been dscussed by few authors, but ts mplementaton for a PUMA robot usng an open archtecture platform was not studed so far. The authors employ the adaptve fuzzy control usng COEM method as a control strategy to mplement the open archtecture desgn for PUMA 56 robot. II. ORIGIAL PUMA UIMATE 5 The Unmaton Mark II s an ndustral robot controller as shown as n Fg.. It conssts of manly a DEC LS seral (Advance onlne publcaton: 2 May 28)

2 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ Fg.2 PUMA robot 52 onts dentfcaton Fg. Unmate Puma 52 block dagram nterface board; CMOS board and EPROM board; servo nterface board and Sx dgtal servo boards [2]. The orgnal system used a large number of operatonal amplfers and dscrete components for condtonng of shaft encoder sgnals and amplfcaton of analog control voltages. Ths leaves consderable scope to smplfy and compact the controller desgn by substtuton of more modern components. III. EW HARDWARE COFIGURATIO The PUMA 52 robot used for work s descrbed n Fg.2. It s a member of the Unmate 56 seres of Robots, havng sx onts. Each of PUMA 52 onts s drven by through a gear tran by a permanent magnet DC servo motor whch ncorporates a rotary shaft encoder, a tachometer and a potentometer. The new system s block dagram s shown n the Fg.3. The PWM amplfer box contans 6 n- house bult amplfers employng SA. The SA amplfer s a pulse wdth modulaton amplfer that can supply 2KW to the load. The control box ncludes an nternally desgned dgtal condtoner card for shaft encoder s sgnals and an analog condtoner card for potentometer and tachometer. The n-house bult encoder condtoner card uses ALTERA MAX 7256AETC- CPLD [3]. It belongs to MAX 7A programmable devce famly. The card has 3 CPLDs one for each shaft encoder The robotc arm needs two dgtal condtoner cards. The CPLDs are programmed usng VHDL language. The sgnals A+, A-, B+, B-, Z+ and Z-, VCC and DGD are the eght sgnals from rotary shaft encoder whch are nterfaced to the CPLD va a dfferental lne recever MC3486. The 24 sgnals D_wast to D23_wast go to 6 channels 722 DIO card. The other 5 onts shaft encoders are connected to dgtal condtoner card n the same way. The power supply unt ncorporates power supples for PWM amplfers unts, sgnal condtoner cards and an exctaton V power supply for 6 servo motors. A Pentum IV ndustral computer s used as a central controller. It has one 6 channel 722 DIO card, 6 bts 86 A/D and 626 D/A cards. The analog sgnals from tachometers and potentometers are fed nto the analog condtoner card. The card was desgned at Smulaton & Machne Control lab (S & MC). After condtonng the sgnals, they are fed to ndustral PC (A/D card). The analog feedback sgnals from D/A are provded to PWM amplfers for each ont to complete the speed loop. A. Advantages of new PC based PUMA robot A new PC-based PUMA robot manpulator control system has three advantages: smplcty, flexblty and low cost. The hardware and software complexty of the Unmate Mark II robot controller s extensvely reduced as dscussed n secton 3. Flexblty refers to the ablty to mplement arbtrary control strateges whch can easly ntegrate sensor nformaton nto low level control. Flexblty also refers to the ablty to easly use wde varety of sensors n the traectory generator. The suggested PC based platform has the ablty to easly ntegrate such sensors as sonars, rangng lasers, cameras etc. that would allow the user to mplement complex control strateges (e.g. vson based control). Wth the current dgtal servo boards n the Mark II, none of these advanced modes are feasble. Another potental advantage of the suggested platform s ts cost effectveness. Due to economes of scale and ncreased CPU power, personal computers have become vable and cost effectve alternatves to workstatons n engneerng applcatons. An Intel processor easly outperforms a Spark IPX whle costs less than a thousand dollars. IV. DESCRIPTIO OF THE COTROL SCHEME A. Manpulator dynamc model The dynamc model of an n-lnk manpulator s gven as [4]: (Advance onlne publcaton: 2 May 28)

3 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ M(q) q& + C(q,q)q & & + G(q) = τ () Where q s the n dmensonal ont varable, q& s the tme dervatve of q,.e. velocty, M(q) s the n n nerta matrx, C ( q, q&) q& s the n-dmensonal vector of corols and centrfugal force, G(q) s the n-dmensonal vector of gravty force and τ s the n dmensonal vector of appled torque. After takng the effects of random dsturbances nto account, the manpulator dynamc model becomes as under: M(q) q& + C(q,q)q & & + G(q) + Bd (q,q) & = τ (2) Where ( q, q& ) represents an addtve bounded B d dsturbance due to load varaton and modelng errors. A. The Controller Output Error Method The Controller output error method s a technque of fne tunng the parameters of a fuzzy system wthn the control archtecture shown n Fg.4.COEM s a strategy for adaptng the parameters of a controller wthout the use of an mplct or explct plant model. The underlyng dea of COEM s as follows. Each tme the response of a plant to a set-pont sgnal s observed, we learn how to repeat that response, should t be requred n future. Tradtonally, adaptve control strateges have been categorzed nto two groups: drect and ndrect. These approaches rely mplctly or explctly on a plant model n order to derve the approprate change n each controller parameter from the plant output error, e y ( = r( = y(. In COEM, the plant output error s not utlzed and therefore no plant model s requred. The lack of dependence on the plant output error, however, ntroduces a restrcton on the applcaton of COEM. Snce the cost functon s not based on the plant output error, ths error s not drectly mnmzed. Ths mples that f the controller does not already stablze the plant, COEM s not lkely to cause t to do so. In other words, the controller must be able to stablze the plant before COEM s utlzed. We consder the non lnear plant y(k + ) = F(y(,...y(k q + ),u(,...u(k q + ) (3) where y ( s the plant output at nstant k F(.) s a non lnear functon and u( s the control sgnal. The constant p and q defnes the order of the plant. At any nstance the state of the plant may be defned by T S = [y(,...y(k p + )] Assumng that the plant s observable). The fuzzy controller produces a control sgnal u (, whch drves the output of the plant to y ( k +). Regardless of whether or not ths was the desred response, f the transton from state S to an output y( k +) s requred agan n future, the approprate control sgnal for the plant would be u (. Ths s n contrast to the normal control problems where the am s to desgn a controller whch wll drve the plant output to track the gven reference sgnal r (. The fuzzy controller n fg.4 outputs the sgnal u(ˆ nstead of producng a control sgnal u(. Thus the output error s e u ( = u( ˆ u(. An mportant pont s that although u(ˆk ) s produced by the controller; t s not appled to the plant. In fact t s used to calculate e u (. u(ˆk ) s computed by producng a new controller nput vector whch s passed through the fuzzy controller. A fuzzy controller can be defned as a functonal relaton between a set of nputs and outputs. Ths s gven by Fg. 3 Schematc dagram of new robot hardware (Advance onlne publcaton: 2 May 28)

4 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ u( = G(z(,Θ( ), where In ths paper the lngustc values are defned by Gaussan membershp functons and are gven as: z( = [r(,y(,...y(k n + ),u(k ),...u(k m)] (4) Where G( ) s the functon defnng the controller characterstcs, Θ( s the set of controller parameters and z( s the controller nput vector and, m and n are the constants. ote that l=m+n+. The fuzzy controller s parameterzed by Θ ( = { µ (,σ (, β (; =,2...; =,2,... l û( s calculated by producng a new controller nput vector, ẑ ( whch s passed through the fuzzy controller as u( ˆ = G(z(,Θ() ˆ, where ˆ T (5) z( = [y(k + ),y(,...y(k n + ),u(k ),...u(k m)] The nput vector z( dffers from ẑ( n the frst element, where y(k+) replaces r(. We have chosen to use the Gaussan membershp functon and the centrc method of fuzzy nference n ths paper as a concrete fuzzy structure to convey the dea. The method to be descrbed n Secton 4.3 s not dependent on the choce of membershp functon nor the choce of nference mechansm. Whether one chooses to use, for example, a trangular membershp functon or the Takag-Sugeno nference mechansm, the method developed n Secton 4.3 can be appled drectly. A. Fuzzy controller system Consder the non lnear plant (3). The Sugeno s type nverse fuzzy model [5] of the process can be formulated as gven below: The fuzzy rule base R contans a set of fuzzy rules as 2 R = { Rule, Rule... Rule } where the th fuzzy rule s ~ Rule : f z( s A then u( s β Where z )] the fuzzy controller, and ~ ~ ~ A = [ A,... Al ] s a lngustc vector referrng to fuzzy varable z (. β s a consequent parameter correspondng to the control sgnal u (.The number of ndvdual membershp functons for specfc nput value z ( s k. T ( = [ z (,... z l ( k s a vector whch contans all nputs to Fg.4 Fuzzy control system usng COEM T ~ (z 2 ( µ /σ ) A = e (6) Where σ and µ are unknown constant parameters. These parameters wll be adusted on-lne usng a Gradent Descent Algorthm. The output of the controller s obtaned as: β u( = = ω β = ω Where represents the consequent parameters and the rule-frng- strength gven as: l = (7) ω s ~ ω = A (z () =,2,... (8) B. Fuzzy control ntalzaton Whle desgnng a fuzzy controller, the lngustc rules are drectly derved from the expert knowledge of the plant or some numercal technques can be used to generate the rules automatcally. The latter approach was utlzed here. The algorthm for ntalzng the controller was developed usng the followng steps: t t ) Generate T numercal data samples ( z,u ) by applyng random sgnals to the nput of the plant. 2) Specfy the maxmum and mnmum values of all + nputs Z, Z to the fuzzy controller. 3) Defne a number of ndvdual membershp functons K for each nput. 4) Determne the ntal centreµ, of each antecedent membershp functon, dstrbuted unformly over the unverse of Z usng an + nterval Z Z /K. The centers are chosen such that they are dstrbuted evenly over the unverse of Z. 5) Determne the dspersons, membershp functon as + σ of the antecedent σ = α Z Z /K, =,2,... (9) where α s the nteracton coeffcent of membershp functons. 6) From the rule-based system wth all the possble combnatons of the membershp functons. The maxmum possble number of rules s K. The shape of the Gaussan membershp = functon wth 6 rules s shown n the Fg. 5. 7) A heurstc method (8) s used to set the parameters of the consequent part as (Advance onlne publcaton: 2 May 28)

5 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ Fg. 5 Membershp functons for fuzzy controller wth 6 rules. β = T l t= = T l t= = ~ A (z )u ~ A (z ) t t t () 8) If the system behavor s stll not desrable, then we have to repeat the controller ntalzaton steps wth more membershp functons and larger nteracton coeffcent. C. Parameter adaptaton usng Gradent Descend Algorthm The controller output error s used n a cost functon such as 2 J( = eu( () 2 The adaptaton of controller parameters s done to mnmze the cost functon. For ths method to be effectve, t s desrable that the membershp functons employed have a contnuous frst dervatve. The Gradent Descend algorthm s employed here. The parameters are updated by: J ( Θ( k + ) = Θ( η Θ( ow usng (), the partal dervatve of the cost functon, J(K), to each parameter can be formulated as J( eu(ω = β ω = 2eu (( Zˆ ( J( = σ( (σ l l [ cω (β u() ˆ ] 2eu (( Zˆ ( (2) J( = u (σ ω µ ) l= 2 ) = l l [ cω (β u() ˆ ] µ ) l= 3 ) ω = V. SOFTWARE DESIG OF THE COTROLLER To mplement the control algorthms developed n secton 3, real-tme software was developed usng C++ and VC++ [5]. Gven the nature of the hardware, the software desgn s based on the use of tmer nterrupts to generate the dscrete tme controller samplng perod. The nterrupt servce routne runs the controllers for all sx onts. There were some mplementaton constrants to optmze the code wth respect to memory and computaton tme for real tme system. To solve ths problem, we have optmzed our software codng to decrease the computaton tme, however, at the expense of memory. At the start of operaton the robot s calbrated. The calbraton procedure begns by runnng real-tme control wth the encoders reset and the demand poston set to. The real-tme controller wll not cause any movement as the demand poston values are equal to the actual values. For each ont n turn, the man program wll check the ont potentometer value and drve the ont toward the home poston. Upon reachng the potentometer home poston and locatng the nearest shaft encoder ndex pulse the correspondng quadrature counter wll be cleared and the desred value set to. The procedure s repeated untl all onts are properly calbrated. The graphcal user nterfaces developed for robot are shown partally n Fg. 6 and Fg. 7. Fg. 6 shows dfferent optons for robot control. The Poston-Control and Rate-Control are used to control the robot 6 onts poston and speed respectvely. The Sgnal-Generator s desgned manly for testng the robot poston-traectory performance. The Data-Vew and data-curve dsplay the onts poston and speed data. Fg. 7 demonstrates Poston-Control wndow only. The software has several levels: System ntalzaton and self dagnoss: whch ntalzes custom boards, confgures the robot and dagnoses each block of the system; System coordnaton and safety check: whch works wth the safety devce to montor robot status and stop operatons n case of errors or emergency; Basc I/O routnes for feedback nformaton and output control sgnals: whch reads ont encoders, poston sgnals, estmate veloctes and convert dgtal control sgnals nto analog ones; Where η s the rate of descend. If the lth rule s dependent on the th membershp functon of the th nput, l c s equal to otherwse equal to zero. Fg.6 GUI of PC based PUMA Robot (Advance onlne publcaton: 2 May 28)

6 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ Knematcs & dynamc routnes: Ths ncludes forward 5cos(5t) and nverse knematcs for path plannng as well Bd = -m 5cos(5t) dynamc routnes. User nterface: Ths provdes users wth control buttons The mplementaton uses 6 rules ntally, wth four to properly operate the robot, convenent means of ndvdual membershp functons for each of y( and r(, plannng experments and post-processng expermental.e. K=4 and K2=4. The nteracton coeffcents are data. α =α 2 =.7 whle η =. was used. The shape of the four VI. RESULTS For ths robot, the Denavt-Hartenberg can be defned n accordance of coordnate system. Matrx A n shows relatve postonng of lnks. cosθn snθncosθ n snθnsnαn a ncosθn snθn cosθncosα n cosθn snαn ansnθn = Where An snα cosα d n n n c s c2 s2 a2c2 s c = = s2 s2 a2s2, A, A 2 d 2 c3 s3 a3c3 c4 s4 s3 c3 a3s3 = = s4 c4, A3,A4 d 3 c s c = s c and 5 5 A s 6 A 5 6 = d 5 n = cos n, sn = snθn for n = s c 6 6 c θ,2,...6 And d n, a n andθ n : Constant D-H parameters. The parameters of the robotc arm manpulators are l=.3m; l2=.44m; l3=.38m; m=25; m2=26.3; m3=5. The frcton and dsturbance terms are taken as: Gaussan membershp functons s shown n the fg The suggested system was developed and appled to Puma 56 robot. The orgnal propretary controller was replaced wth t. The PC runs the Wndows 2 operatng system. The PC-based controller s evaluated from two dfferent aspects. The frst aspect s to examne how easy the system ntegratons and modfcatons are. The second s to examne the performance of the control. The frst aspect s obvous. The suggested PC-based hardware and software ensure that the extensblty and scalablty are avalable. The second aspect s evaluated by the traectory-trackng experment. To verfy the effectveness of the new controller, experments were performed to test the trackng control of the robot manpulator. Frstly, each ont s separately requested to follow a desred traectory. In ths test, each ont s asked to move to a specfed destnaton whle followng a predetermned path. Fg. 8 and Fg. 9 show the desred poston traectores and poston trackng errors respectvely for sx onts. The desred profle conssts of two parts: (a) lnear acceleraton from rest to maxmum velocty for onts, 2, 3 and 5 whle lnear deceleraton for ont 4 and (b) decelerate lnearly. The poston trackng errors of ont, 3, 4, 5 and 6 are n the acceptable range of.5rad to.rad except ont 2 for whch the trackng error s.2rad. To further verfy the controller traectory trackng performance, all the onts were requested to follow a varyng Fg.7 Robot Poston-control GUI layout (Advance onlne publcaton: 2 May 28)

7 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ Fg 8 Desred poston traectores (Radans vs. Seconds) Fg 9 Poston trackng error (Radans vs. Seconds) Fg. Traectory trackng error for ont. (Radans vs. Seconds) The same tests were performed wth varyng onts veloctes and acceleratons. Jont and ont 3 showed hgher poston trackng errors at hgher veloctes and acceleratons, however, all the remanng onts showed satsfactory performance. All the onts show satsfactory performance at low velocty as well as they exhbt low poston trackng error whle followng a velocty profle at hgh speeds. For robustness, The FLC s compared wth the conventonal PD controller. The PD controller was frst well tuned for a normal operaton. Then frcton and dsturbance was ntroduced n the hardware testng. The results n Fg.2 to Fg.5 show that the proposed controller for the PUMA robot s much more robust as compared to a well tuned PD controller. traectory. Fg & Fg. shows the ont traectory trackng performance and error respectvely. All the other onts show acceptable trackng performances. Two further tests were performed to examne the smultaneous onts movement: () the maxmum velocty was set at 5 counts/sec. To test the smultaneous onts movement, all sx onts were asked to move at ther fastest respectve speeds and (2) the lnear nterpolaton mode test n whch the ndvdual onts should arrve at ther respectve destnatons at the same tme. Fg. 2 FLC wth frcton.23m (Jont of robot) (Radans vs. Seconds) Fg.. Comparson between planned and followed traectory of ont (Radans vs. Seconds) Fg. 3 PD wth frcton.23m (Jont of robot) (Radans vs. Seconds) (Advance onlne publcaton: 2 May 28)

8 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ controller from off-the-shell hardware and software components, the benefts of reduced and mproved robustness have been realzed. Although the experments were performed at educatonal and research nsttuton, the research s orented towards ndustral applcatons The paper uses the method Controller Output Error method (COEM) for on-lne tunng of the fuzzy controller. Ths method can be used wth any fuzzy controller desgn; the only requrement s that the controller must be stable Fg. 4 FLC wth unt nput dsturbance before ntalzng the tunng. So any fuzzy rule-based model (Radans vs. Seconds) can be used to parameterze and ntalze the controller of the system. COEM s subsequently appled to tune the parameters of fuzzy controller for purpose of achevng better performance. Gaussan membershp functon and the centrc method of fuzzy nference n ths paper are chosen as a concrete fuzzy structure to convey the dea. The method descrbed n Secton 4.3 s not dependent on the choce of membershp functon nor the choce of nference mechansm. Whether one chooses to use, for example, a trangular membershp functon or the Takag-Sugeno nference mechansm, the method developed n Secton 4.3 can be appled drectly. Fg. 5 PD wth unt nput dsturbance (Radans vs. Seconds) Though, some techncal problems were faced whle performng the tests at hgher veloctes for ont and ont Another experment was done to compare the performance of 3, the newly desgned hardware and software works very well our mplemented fuzzy controller wth LQR and sldng and overcomes the problems n the prevous PC based desgn mode controller. Though the other controller also show good for PUMA robot. All the onts show satsfactory trackng performance, FLC was able to compensate for some performance at low velocty as well as they exhbt low dsturbances and sensng nose as shown n the Fg. 6(a, b poston trackng error whle followng a velocty profle at and c). hgh speeds. The expermental results showed that t s feasble to VII. DISCUSSIO AD COCLUSIO mplement modern control methods for PUMA 5 Seres In ths paper, the prelmnary results are acheved n the Robots through software routnes runnng on a PC. Presently the authors are workng to mplement and adapt development and mplementaton of a new smple PC based the system for an ndustral CC mllng machne. replacement controller for PUMA 52 robot. By assemblng (a) ( ( (b) (c). Fg.6.Robot Jont trackng performance (a) FLC mplementaton, (b) LQR mplantaton, (c) Slde mode control mplementaton (Advance onlne publcaton: 2 May 28)

9 IAEG Internatonal Journal of Computer Scence, 35:2, IJCS_35_2_ ACKOWLEDGMET The authors would lke to thank to Mss Ba Tngtng who helped us n robot hardware desgn and Mr. Ye Guoha for hs contrbuton n mechancal desgn and assembly. We would also lke to thank Dr. Huang for helpng us n debuggng the GUI software n VC++. REFERECES [] S. Sorenson, Overvew of Modular, Industry standard based open archtecture controller, Proc. Intl. Conference of Robots and Vson Automaton, Detrot Mchgan,, 993, USA, [2] D. J. Mller, R. C. Lenox, An Obect Orented Envronment for Robot system Archtecture, Control system magazne, IEEE, 99, Vol, Issue 2, pp [3]. M. B. Jr. Leahy, S. B. Petrosk, Unfed Telerobotc Archtecture Proect status report, IEEE Intl. Conference Systems, man and Cybernetcs, San Antono, Texas,994, Vol., pp [4]. Unmaton Robotcs. User s Gude to VAL 398P2A: A Robot Programmng and Control System, Unmaton Inc., Danbury, Connectcut, 983, USA. [5]. R. Vstness, Breakng away from VAL. Techncal Report, Unmaton Inc. Danbury, Connectcut, 982, USA. [6]. Pan Langdong, Huang Xnhan., Implementaton of a PC-based robot controller wth open archtecture, Proceedngs. IEEE Internatonal Conference on Robotcs and Bometrcs, 24, pp [7]. V. M. Becerra, C.. J. Cage, W. S. Harwn, P. M. Sharkey, Hardware retroft and computed torque control of a Puma 56 Robot updatng an ndustral manpulator. IEEE Control Systems Magazne, 24, v 24, no. 5, pp [8]. P. Fedler, & C. Schlb, Open Archtecture Robot Controllers and Workcell Integraton, Robotcs Today, 998, Vol., no.4, pp. -4. [9]. J. Katuptya, R. Radaewsk, J. Sanderson, M. Tordon, Implementaton of a PC based controller for a PUMA Robot, Proc. IEEE Conference on Mechatroncs and Machne Vson n Practce, 997, Australa, pp []. H. C. Anderson, A. Loft and A. C. Tso, A new approach to adaptve fuzzy control: The controller output error method. Proc. IEEE conference on system, man and Cybernetcs [C]. Part B. 997, 27(4), 686~69 []. V. M. Becerra, S. Cook, J. Deng, Predctve Computed-torque control of a PUMA 56 Manpulator robot. Proc. 6 th IFAC world congress, 25, Prague. [2]. 5. P. I. Corke, Operatonal Detals of the Unmaton Puma Servo System, Report, CSIRO Dvson of Manufacturng Technology, 993, Australa. [3]. L. Hanho, G. E. Sobelman, VLSI Desgn of Dgt-Seral FPGA Archtecture, Journal of Crcuts, Systems, and Computers, 24, Vol. 3, o. [4]. F. L. Lews, C. T. Abdallah, & D. M. Dawson, Robot Manpulator Control, Theory and Practce, 2 nd ed., Marcel Dekker, 24. [5]. K. M. Passno.and S. Yurkovch. Fuzzy Control. Menlo Park, Addson and Wesley Longman, 998, 25~39. (Advance onlne publcaton: 2 May 28)