Evolution-based virtual training in extracting fuzzy knowledge for deburring tasks

S.-F. Su, T.-J. Horng, K.-Y. Young
Proceedings 2000 ICRA. Millennium Conference. IEEE International Conference on Robotics and Automation. Symposia Proceedings (Cat. No.00CH37065)  
Dcl);~rtnici~t of Elcctriciil ;inti Control Eiigiiiccring, Niitiotliil Ti1i\vi111 Ut~i\~crsity of' Sciciicc i111t1 Tcc111iology T;I~\V;III It. 0. C. Ahstract In this research. tlie probleins oT Iiow to tcacli a robot to execute skilled operations arc studied. Huiiiaii \\,orkcr:s usually accumulate his esperieiicc ;iftcr czecuting the same task repetiti\.ely. I n tlic process of twining. tlic worker must find \\'a!'s of ad.iustiiig Iiidlier execution. In our system. the parametcrs for tlic
more » ... tcrs for tlic iiiipeclaiicc control sclietiie are used as the targets for ;id.iustnicnt. Aftcr ni;iss aiiiount of training. [lie workcr is siipposcd to bc ablc to execute deburring tasks succcssliill! . This is bccriusc tlic worker might 1iai.e gotten sonic kiio\vlcdge ;iboiit tiuiiiig the parameters requircd in tlic iiiipct1;iiicc coiitrol sclicnic. Tlius, tlie ntles for adjusting tlic praiiietcrs i n iinpcdiiiicc control are the operational skills to bc idcntificd. III this research. a training sclicnic. c;illcd tlic c\~olutioii-b;iscd virtual training sclietne. is proposcd i n extracting knowledge for robotic debiirring t;islts. I n this nppro;icli. a evolution strategy is einploycd 10 scircliing for thc bcsc set of fiizq niles. This 1e;irniiig sclicnic lias becii successfiilly applicd in adjust iiig the paraiiictcrs of impedance controllers requircd i n dcburring opcratioiis. In general. the results of dcburring ;ire niucli satisf:ictoi> when compared with tliosc in tlic prc\,ioos rcse;ircli. When executing a deburring task. tlic robot siiiiuliitor caii find its optimal adjusting rules Tor p;iranictcrs aftcr several generations of e\,olutiori. In t rod ti ct ion Recently. industr?; lias succcssfiill!. uscd robots in engaging in executing various t:isks \\.liosc \\.orkiiig environment is hartnful to 1i1111i;iii bciiigs or wliosc operations are repetitive and/or rcquirc Iiigli iiccur:ic!~. Usually. those tasks can bc progriiiiinicd into tlic operations of robots because tliosc i;isks do not iiitcfiict with tlie environinent frequently :iritl tlicn liiininn skill may not be necessary for the 0pcr;itioiis of tlic tiisks. 0 1 1 tlie other hand. there exist t:isks. sucli iis dcburring. grinding. milling. assembly. ctc.. \\.liicli iiiay iiccd ;I grcar deal of interactions wit11 the cii\,iroiiiiiciit and thus. require lots of decision-making processcs udiilc f;iciiig those interactions. Hcnce. tlic succcssful csccution of those tasks largely relies 011 1iuiii:iii skill i i i ;icliic\,ing satisfactoiy results. Such kinds of asks iirc \.cl?. dillicult to be satisfactorily prograniiiicd into tlic opcratioii of robots. 111 fact. eveii thcrc exists soiiic work t l u t I i x tried to manage to embed tliosc tasks iiiro tlic OpcriiIioIi of robots. those operations iiia!. face lots of problems when unccrtainty occiirs in the cn\ironnient and tlie actual opcriitions ma!. not be satisf;ictoi-y. Sc\cral researchers I i a \~ tried to discover tlie rclutionsliips bchvecn huiiiaii espei-ts' intentions and opcrational stratcgics for tasks so that the skills could be iiiodclcd accordingly and tlicn ;irc possibly transferred to tlie opcrations of robots. As;id;i ef d . have tried to use ncural nct\voi-ks 11.21. ad;ipti\:c control 131, or fuzzy rules 141 to iiiodcl and to trarisfcr l i u n i m skills to the operations of robots. I n 151. i n ordcr to acquire liunian skills. tlic aulliors I i a \~ relascd tlic joints of a robot iiianipulator. and let a 1i~iiii;iii cspen worker take the end-cffcctor of n iiianipu1:itor to accomplish a compliant task. Tor instance. deburriiig. Tlic d m of the deburring proccss. sucli ;is positions (angles) ;ind forces (torques) of a l l .joiiits are rccordcd. The ;ippro:icli is tlicn to extract useful rulcs or stratcgics lor rcprcscnling tlie skills from 111c collected data. Ho\\-e\.cr. sonic problems inay arise in the above approach. First. tlic obtained niies or strategies basccl on tliis sct of data niay not bc able to represent tlie skills in sufficiency. When a n cspcrt ivorker need to take thc aid-effcctor to proccccl the esccution. since in this unusua.l \\'ay tlic \vorkcr c;iiiiiot execute the task as he/she usu;illy did. tlic rcsults nia!iioi be satisf:ictory. Besides, tlic rccorded data Iiiiiy not bc sufficient to adequately csprcss tlic opcrational sltills. Secondly. the rules or strategics cstractcd froiii tlic c1;ita arc rather primitive and arc scnsiti\t to tlic oper:ition;il coiiclitioiis. For example, i n dcburring tasks. if difrcrclll iiiatcrial properties or surfiicc rougliness of wrkpicccs arc considered for deburring tasks. tlie obtaiiicd stratcgies inay not work-\vcll. Thercrorc. it may bc rcquirecl lo filnlier generalize tli;it obtained kno\vlcdge to copc \villi the variation of tasks. Tlius. in this rcscilrcli \ I C attempted to propose anotlicr \\ay of dcfining opcriitional skills and training schciiics for Icariiing tlic dcburring tasks. III tlic work of 11 1 ). Iiiiriiaii skills ;ire stored in the clcsircd positioii conini;iiids to tlic controller of a robot ni;inipul;itor. I n that >il)l)roiicli. otlicr piirameters, for dcburriiig tasks. the ~~;iranictcrs rcquircd in the iiiipcd;incc coiitrol schcinc arc choscn as constants. Ho\\.c\.cr. tlic rcsults in o w siniul;itiori have shown that the perforinancc can bc iiiipro\.cd by changing the p;ir;inictcrs i n tlic inipcc1;iiicc control mechanism. In ordcr IO obtain bcttcr rcsults for dcburring operations, a w:iy of dctcniiiiiiiig \vIicii ;incl IIOW to change those
doi:10.1109/robot.2000.845332 dblp:conf/icra/SuHY00 fatcat:btn2ew25inazni3zy3do7dwwti