Neuro-fuzzy architecture for identification and tracking control of a robot manipulator

Ncural networks and fuzzy systcnis havc bccn applicd very succcssfiilly iii tlic idcntification and control of dynamic systcms. This papcr prcscnts combination of the fuzzy logic controllcr and ncural nctwork identification structurc, intcgra~cd into robotic systcm, to providc cxtcnsivc capabilitics. Wc first discuss tiic fuzzy logic controllcr (FLC), dcscribc its maiii compoticnts such as fuzzificr, fuzzy rule basc, fuzzy infcrcncc cnginc and dcfuzzilicr. Wc then lbcus on thc ncural nctwork (NN) plant modcl, traincd on-linc using tllc backpropagation training optimization algorithni with an adaptivc tcamjiig ntc. Thc optimization algorithm is pcrformcd at cadi satnplc time to computc thc optimat control input. The rcsults confirm thc cffcctivcricss o f the proposcd idcoti lication and control arcliitccturcs. Robotic manipulator systcrns arc notihncar, high couplcd, and timc varying. Robots havc to lac, many unccrtaintics in thcir dynamics, in particular structurcd unccrtaintics, which arc causcd by iniprccisioo in the manipulator link propcrlics. unknown loads, and unstructurcd otic, such as nonlincar friction, disturbanccs, and thc high-frcqucncy part of the dynaniics [ I]. Tlic coiitrol pcrforniancc of thc robotic manipulator is inlliicnccd by lhc mcntioncd unccrtaintics of the plant. A convcntional approach to solvc tlic robotic control problem is to iisc [tic coniputcd torqiic algorithm 121. Tlic computcd torque algorithm amounts to tranafomling tlic highly nonlincar robot dynamics into cquivalcnt Lincar systcm. Thcn lincar control tlicory can bc applicd to synthcsizc thc controllcr to mcct the dcsircd specifications. Thc theory of furzy and ncural control S C C ~ S to bc a suitablc tool for both modclling and control of coinpicx, nonlincar systcnis. Fusion fuzzy systcms and ncural nctwork providcs human-like knowlcdgc processing capabilitics. Thc using of FLC for controlling a robot manipulator is justificd liom Ihc followiiig rcasons: thc dynamics of robot is niodclcd by nonlincnr and couplcd diCfcrcntiaI cquations and FLC givcs high flcxibility, that is it lias many dcgrccs 01' liccdoin (shapc and number of mcnibcrship functions, aggregation mcthods, fuzzilkation and dcrtizzification mcthods, ctc.). Fuzzy systcms arc suitnblc for uncertain and approximatc rcasoning, cspccially for tlrc systcm with a miithcmntical iiiodcl that is diflicult to dcrivc [I], [3]-[ 5 ] .