23 7 200 0 ROBO T V o l. 23,N o. 7 O ct., 200 : 00220446 (200) 0720605204 α ( 200030) :.,. : ; ; ; : T P24 : B APPL ICAT ION RESEARCH OF ART IF IC IAL NEURAL NETWORK IN ROBOT TRAJECTORY PLANN ING L IU Cheng2lang ZHAN G Ka FU Zhuang CAO Q 2x n Y IN Yue2hong (R esearch Insttute of R obotcs S hangha J aotong U nversty, S hangha 200030) Abstract: T he app lcaton of artfcal neural netw o r n robo t trajecto ry p lannng s studed. T he penalze func2 ton m ethod s used n co llson free trajecto ry p lannng and the BP algo rthm s gven n avo dng co llson. A t the last, the computer sm ulaton s p rovded. Keywords: ANN, robo t, trajecto ry p lannng, computer sm ulaton ( In troducton). 40, 943 W. ecu lloch W. P tts P. 957 Ro senb latt Percep tron,. 60 W dron. 80 Hopfeld. 986 R um elhart cclelland BP,,,.,.,,. : [ ] [2 3 ] [4 5 ],,., ;,,,. 2 (Expresson of free-coll - s on tra jectory),.,. NN. :,, α : 200-08- 25
606,. NN F g. NN structure of penalty functon fo r obstacle x y z,, x y z.., 0. 5. NN I m : C f (T I) T I O m m I + ΗT O m f (I m ) w xm x + w ym y + w zm z + Η m C ; T I ; ΗT ; O m I m Η m w xm w ym w zm m, S f (x ) + e -. x gt, ;. 3 (Tra jectory plann ng of po n ted object) : ;.,,. N K C E c K : ; C : P.. : P (x, y, z ),, 2,, N : N - N - E l L 2 [ (x + - x ) 2 + (y + - y ) 2 + (z + - z ) 2 ] L : ; E l:. : E w le l+ w ce c; w l, w c. E,. x α - y α - z α - w l + w l + w l E α (g p E ) T p α 5L 2 5 x E α - + 5L 2-5 x 5L 2 5 y 5L 2 5 z + 5L 2-5 y + 5L 2-5 z 5L 2 + 5L 2-5 x 5 x 5L 2 5 y 5L 2 5 z + 5L 2-5 y + 5L 2-5 z 5 C 5 x x α 5 C 5 y 5 C 5 z y α z α 5 C 5 x 5 C 5 y Γ (x α2 + y α2 + z α2 ) < 0 5 C 5 z x α2 0, y α2 0, z α2 0 E α 0, E.. 5L 2 + 5L 2-2 - 2x + + 4x - 2x - 5 x 5 x 5 C 5 x 5 C 5 (T I) 5 C 5 (T I) 5 (O m ) 5 (I m ) m 5 (T I) 5 x 5 (I m ) 5 x f [ (T I) f [ (I m ) ]w xm m P (x y z )
23 7 : 607 x α - Γ[ 2w l (2x - x - - x + + w c f [ (T I) m f [ (T m ) ]w xm y α - Γ[ 2w l (2y - y - - y + + w c f [ (T I) m f [ (T m ) ]w ym z α - Γ[ 2w l (2z - z - - z + + w c f [ (T I) m f [ (T m ) ]w zm f (g) f (. ) [ - f (. ) ] T,. 4 (Com puter sm ula ton) 4. 2, 4 : x - 0. 2 > 0 - x + 0. 8 > 0 y - 0. 2 > 0 - y + 0. 7 > 0 3. 4 T. 4, T, 2 F g. 2 R ectangle obstacle 3 F g. 3 N eural netw o r 4 F g. 4 3D shape fo r co llson penalty functon T,, 0.
608,,. T. ; :,,,. T T. T (t) T 0 log ( + t) T (t) : T 0 T, + t,., T,,. 4. 2,, 5,. 6, 5. 5 F g. 5 Co llson free trajecto ry p lannng,. (References) 6 F g. 6 Co llson penalty functon 3D shape 5 (Concluson) ANN,, NN,.,,.., 992, 4 (6) 2 Behera L, Chaudhury S, Gopal. A pp lcaton of self- o rganz2 ng neural netwo r s n robo t trac ng contro l. IEE P roceedngs: Contro l T heo ry and A pp lcatons v 45 n 2 ar 998 IEE Steve2 nage Engl, 35-40 3 3 Smon, D an. A pp lcaton of neural netwo r s to op tm al robo t trajecto ry p lannng. Robo tcs and A utonomous System s, 993, () : 23-34 4 5 Smon, D an. N eural netwo r2based robo t trajecto ry genera2 ton. IEEE Internatonal Conference on N eural N etwo rs ar 28 - A p r 993: 540-545 5 8 acuow. Robo t contro l w th neural netwo r ṡ A I and In2 fo rm aton Contro l System s of Robo ts, P lander. N o rth- Ho lland, 989 (. )
604.,,., 360., 5. 4. :, a b c d e f g hh h j l m no p qrs t u b 4 IV ECO F g 4 T he trace p lannng of IV ECO cro ssbeam 2 IV ECO F g. 2 V IECO cro ssbeam 4. 2 5 F g. 5 trace of w eldng header on cartesan coo rdnate (References) Ish hara, Ko ch. N ow and future of laser w eldng robo ts 2 app l2 caton n automo tve ndustry. Robo t n 4 Jan 997 Toyo Japan p 42-48 2 Yam aoa, N ao j; O da, Kouj. O n2lne car2body m easurng system usng modfed car2body w eldng robo ṫ Robo t n 4 Jan 997 Toyo Japan p 77-8 3 H arw g, D enns D. W eld param eter developm ent fo r robo t w eld2 ng Techncal Paper2Socety of anufacturng, RP P roceedngs of the 996 anufacturng Conference Sep 4-996, P96-29 996 D earbo rn I U SA 3 IV ECO F g. 3 T he w eldng trace of IV ECO automoble cro ssbeam : (9642),,,. :, CAD,. 40, 2, 3..