2 Trvizan, Vloo(2010) RoboCup 2 [3] 22 Epiod 2 Epiod Vir, Wland(2003) RoboCup [4] O x ( ) x π/2 y t 0 m R i (t 0 ) (1 i m), t 1 n R j (t 1 ) (1

Size: px

Start display at page:

Download "2 Trvizan, Vloo(2010) RoboCup 2 [3] 22 Epiod 2 Epiod Vir, Wland(2003) RoboCup [4] O x ( ) x π/2 y t 0 m R i (t 0 ) (1 i m), t 1 n R j (t 1 ) (1"

こごろうやすこ
5 years ago
Views:

1 社団法人人工知能学会 Japan Socity for Artificial Intllignc 人工知能学会研究会資料 JSAI Tchnical Rport SIG-Challng-B30 (5/5) RoboCup Analyzing and Larning Opponnt Stratgi in th RoboCup Small Siz Lagu Kotaro Yaui, Kunikazu Kobayahi, Kazuhito Murakami and Tadahi Naru Graduat School of Information Scinc and Tchnology, Aichi Prfctural Univrity {koboyahi, murakami, Abtract In th RoboCup Soccr w can dominat a gam by chooing an ffctiv tratgy if w can analyz and larn opponnt tratgi in advanc. Solving thi problm i a challnging tak, o w attack th problm in thi papr. W propo a diimilarity function which how th diffrnc btwn opponnt dploymnt at two diffrnt tim, and xtnd it to th diffrnc btwn tho of two diffrnt tim intrval. Thn, w analyz opponnt tratgi by uing th diimilarity function. A a firt tp w try to claify th opponnt tratgi ud in t play. Employing th diimilarity matrix gnratd from th diimilarity function, w tak th clutr analyi and claify th opponnt tratgi. W apply thi mthod to th loggd data of th mall iz lagu gam playd in RoboCup By th xprimnt, w how w can ffctivly claify th attacking tratgi ud in t play. W alo dicu a mthod to larn th opponnt attacking tratgi and to dploy th tammat in advantagou poition on-lin in actual gam. 1 RoboCup (RoboCup Small Siz Lagu) 1 6 6m 4m Rfr Box Rfr Box m/ [1] 4m/ RoboCup d d 2 Bowling (2004) RoboCup [2] 1

2 2 Trvizan, Vloo(2010) RoboCup 2 [3] 22 Epiod 2 Epiod Vir, Wland(2003) RoboCup [4] O x ( ) x π/2 y t 0 m R i (t 0 ) (1 i m), t 1 n R j (t 1 ) (1 j n) t 0 t 1 d d(t 0, t 1 ) = { } min min trac(f (U) Pσ ) U {U 1,U 2 } σ S 6 [ ] 1 0 U 1 =, U 2 = 0 1 F (U) = [f ij ] [ ] (1) U R i (t 0 ) R j (t 1 ) 2 (1 i m and 1 j n) f ij = 2 (othrwi) (2) S 6 6 P σ σ U 2 y x F 6*6 t 0 t 1 (2) 2 d = 0 (1) t 0 t 1 1 x 1 1 U d d d (1) t 0 T T d 1 d 1 (t 0, T, T ) = (3) T (i) X i T (j) T (j) d d 2 (T (i), T (i), T (j), T (j) min {d(t 0, t)} (3) T t T T (i) X j ) = min T (i) t T (i) {d 1 (t, T (j), T (j) )} (4) 2 (4) (4) 4 N (4) N N 1 2

3 3 4.1 k-man k k [8] Ward 2 [9] k-man Ward 4.2 Davi- Bouldin indx (DBI)[5] DBI K DB(K) DB(K) = 1 K K i=1 max j i S i + S j M ij (5) x (i) C i M ij C i, C j (Sparation) S i C i (Cohion) M ij S i ( [5]) DB(K) K DB(K) 0 (4) M ij S i S i = M ij = 1 C i ( C i 1) { d 2 (T (k), T (k), T (l), T (l) X l C i,x l X k X k C i 1 C i C j X k C i ) d 2 (T (k), T (k), T (l), T (l) X l C j ) S i M ij 2 C i, C j S i, M ij [5] 5 RoboCup ( ) X i i (1 i N) X i T r (i) (Rfr Box ), X i T (i) T (i) = max(t (i) (4) d 2 (T (i) } T bhavior, T (i) r ) (6), T (i), T (j), T (j) ), (1 i N and 1 j N) N N (5) K K T bhavior T bhavior T bhavior T bhavior = 1.0c (5) K [6] 1 K log 2 N + 1 (7) x x x 3

4 4 5.1 RoboDragon RoboDragon A i, (1 i 4) RoboDragon RoboDragon (Blu) RoboDragon (Yllow) 6 Yllow y 2012 [7] 1915mm x 4 6 Fig.1 DBI Fig.2 3 d X23 X24 X19 X20 X21 X22 X3 X6 X5 X4 X1 X2 X10 X11 X12 X7 X8 X9 X15 X13 X14 Figur 1: Dndrogram (RoboDragon) X18 X16 X17 Fig.1 K = 5 5 C 1 = {X 1, X 2, X 3, X 4, X 5, X 6 } C 2 = {X 7, X 8, X 9, X 10, X 11, X 12 } C 3 = {X 13, X 14, X 15 } C 4 = {X 16, X 17, X 18 } C 5 = {X 19, X 20, X 21, X 22, X 23, X 24 } 4 6 A 1, A 2, A 4 C 1, C 2, C 5 A 3 2 C 3, C 4 K = 4 C 3, C 4 Fig.1 C 5 A Skuba(Blu) ZJUNlict(Yllow) Skuba, ZJUNlict, Skuba 37 ZJUNlict 25 Fig.3, 4 d X21 X27 X29 X2 X3 X5 X16 X22 X18 X37 X8 X31 X17 X25 X26 X10 X14 X30 X4 X7 X1 X6 X13 X33 X9 X19 X36 X28 X34 X24 X11 X20 X32 X35 X15 X23 X12 Figur 2: Davi-Bouldin indx (RoboDragon) Figur 3: Dndrogram (Skuba) Fig.2 K = R (5) Skuba K = 5, ZJUNlict K = 6 i X i T (i) k 2 4

5 5 6 d X15 X14 X25 X16 X6 X24 X12 X18 X4 X21 X20 X3 X11 X9 X5 X19 X1 X2 X10 X23 X8 Figur 4: Dndrogram (ZJUNlict) X13 X17 X7 X22 5 N + 1 N 5 C i X j t d 3 d 3 (t, C i ) = 1 C i d 1 (t, T r (j) X j C i, T (j) ) (8) 3 5 ID Skuba X 17 ZJUNlict X 16 x Skuba C 1 C 2 C 3 C 4 C 1 C 5 ZJUNlict C 1 C 2 C 3 C 1 C 4 C 5 C 4 C 6 C 4 2 t C i X j (8) (8) (6) T (j) T r (j) (8) Skuba 37 X X 1 X Skuba C 2 X 37 Fig.5 X 37 Skuba d3 ( t, Ci ) t cond bfor kicking C1 C2 C3 C4 C5 Figur 5: (8) Rfr Box 4 C 1, C 4 d 3 (t, C i ) 5

6 6 Skuba(Blu) C1 = X10, {X1, X4, X6, X7, X14, X17, X25, X26, {X2, X3, X5, X21, X22, X27, X29 } {X8, X15, X18, X31, {X9, X11, X13, C2 = X30 } X16, C3 = C4 = X37 } X19, X20, X23, X24, X28, X32, X33, X34, X35, X36 } 6

7 7 C5 = {X12 } ZJUNlict(Yllow) C1 = {X1, X2, X10, X19, {X3, X4, X6, C2 = X23 } X11, X12, X14, X15, X16, X18, X20, X21, X24, X25 } {X7, X13, C3 = C4 = {X5 } C5 = X22 } X17, {X8 } C6 = {X9 } 7

8 4 3 2 1 2 C 2, C 3 C 2 2 Skuba ID:3 1 ID:8 C 3 d 3 (t, C i ) Rfr Box (8) C 3 C 3 ID:8 2 ID:3 ID:8 2 4 7 2 d d [1] Thanakorn Panyapiang, Krit Chaio, Kanjanapan Sukvichai and Phawat Lrtariyaakchai,

8 C 2, C 3 C 2 2 Skuba ID:3 1 ID:8 C 3 d 3 (t, C i ) Rfr Box (8) C 3 C 3 ID:8 2 ID:3 ID: d d [1] Thanakorn Panyapiang, Krit Chaio, Kanjanapan Sukvichai and Phawat Lrtariyaakchai, Skuba 2012 Extndd Tam Dcription, 2012 [2] Michal Bowling, Brtt Browning and Manula M. Vloo, Play a Effctiv Multiagnt Plan Enabling Opponnt-Adaptiv Play Slction, Intrnational Confrnc on Automatd Planning and Schduling, 2004 [3] Flip W. Trvizan and Manula M. Vloo, Larning Opponnt Stratgi In th RoboCup Small Siz Lagu, Intrnational Confrnc on Autonomou Agnt and Multi-Agnt Sytm, Springr, 2010 [4] Ubbo Vir and Han-Gorg Wland, Uing Onlin Larning to Analyz th Opponnt Bhavior, RoboCup 2002: Robot Soccr World Cup VI, pp.78-93, Springr, 2003 [5] David L. Davi and Donald W. Bouldin, A Clutr Sparation Maur, IEEE Tranaction on Pattrn Analyi and Machin Intllignc, PAMI-1(2), pp , 1979 [6] Hrbrt A. Sturg, Th Choic of a Cla Intrval, Journal of th Amrican Statitical Aociation, Vol.21, No.153, pp.65-66, 1926 [7] Law of th RoboCup Small Siz Lagu 2012, viwd April 4th 2013, ku.ac.th/_mdia/rul:l-rul-2012.pdf [8] k-man clutring, viwd April 4th 2013, http: //n.wikipdia.org/wiki/k-man_clutring [9] Ward mthod, viwd April 4th 2013, n.wikipdia.org/wiki/ward%27_mthod 8

1 Table 1: Identification by color of voxel Voxel Mode of expression Nothing Other 1 Orange 2 Blue 3 Yellow 4 SSL Humanoid SSL-Vision 3 3 [, 21] 8 325

1 Table 1: Identification by color of voxel Voxel Mode of expression Nothing Other 1 Orange 2 Blue 3 Yellow 4 SSL Humanoid SSL-Vision 3 3 [, 21] 8 325 社団法人人工知能学会 Japanese Society for Artificial Intelligence 人工知能学会研究会資料 JSAI Technical Report SIG-Challenge-B3 (5/5) RoboCup SSL Humanoid A Proposal and its Application of Color Voxel Server for RoboCup SSL