AI

Size: px

Start display at page:

Download "AI"

そうすけふじつぐ
5 years ago
Views:

1 JAIST Reposi Title プレイヤの意図や価値観を学習し行動選択するチームプレイ AI の構成 Author(s) 吉谷, 慧 Citation Issue Date Type Thesis or Dissertation Text version author URL Rights Description Supervisor: 池田心, 情報科学研究科, 修士 Japan Advanced Institute of Science and

2 AI

3 AI : Copyright c 2013 by Yoshigai Kei 2

4 AI AI AI 1997 Deep Blue AI FPS RTS AI AI AI AI AI AI AI AI AI AI RPG

5 AI AI AI i

6 ii

7 1 AI AI AI AI AI AI Sander,B. [1] QuakeIII AI AI Aswin,T.A. [2], Non Player Character Kenneth O.S. [3] AI AI RPG 1

8 2 AI AI AI AI AI 4 1. AI AI AI 2

9 3 RPG [4] D : 3

10 Kahneman [5] 1. A 80% 4, 000 B 100% 3, C 20% 4, 000 D 25% 3, E 45% 6, 000 F 90% 3, G 0.1% 6, 000 H 0.2% 3, B 2 C 1% F 4 G % 80% 25% 20% 3 4 4

11 4 4.1 AI RPG HP 0 MP HP MP αβ 4.2 a 1 a 2 AI b 1 b 2 AI 5

4.1: RPG 4.2 AI AI AI a 1 AI a 2 4.3 AI AI AI AI 4.

12 4.1: RPG 4.2 AI AI AI a 1 AI a AI AI AI AI 4.1 a 1 a 3 AI b 1 b 3 (a 1, b 3 ) AI b 3 (a 2, b 2 ) (a 1, b 3 ) AI b (a 2, b 3 ) (a 1, b 2 ) AI 6

13 4.2: AI b 1 b 1 b 2 b 3 a 1 (a 3, b 1 ) a 3 a 2 a 3 AI 2 1 AI 4.1: b 1 b 2 b 3 a a a

14 5 Markov Decision Process HP 0 0 MP 5.2 8

15 [] HP

16 5.1: 5.1: 5.1 HP MP RPG RPG 10

17 6 AI AI 6.1 AI s a Eu 6.1 Eu(s, u, a) = s {u(s )T r(s, s, a)} (6.1) u T r(s, s, a) s a s b 6.2 Eu(s, u, a, b) = s {u(s )T r(s, s, a, b)} (6.2) 6.2 d i = (s i, a i ) D( d i ) d i s i a i 4 11

18 6.1: AI 1. s A a n A B b m B (a n, b m ) 6.2 EuT able 6.3 EuT able(u, s) = {Eu(s, u, a n, b m )} n,m (6.3) 4.1 EuT able(u, s) 4.2 EuT able(u, s) s i EuT able(u, s i ) EuT able(u, s i ) EuT ables(u, D) = {EuT able(u, s i )} i (6.4) 2. EuT able(u, s i ) s i a i p(s i, u, a i, EuT able) p(s i, u, a i, EuT able) AI p

19 3. u 6.5 loglikelihood(u, D) = i log{ϵ + p(s i, u, a i, EuT able)} i (6.5) ϵ(> 0) p(s i, u, a i, EuT able) u u loglikelihood(u, D) > loglikelihood(u, D) u D u Remi BTMM [] [] 6.3 u t s t EuT able(u t, s t ) EuT able(u t, s t ) 4 AI

20 7 AI AI AI AI AI AI 100 AI AI AI AI AI : 1 2 HP MP AI AI AI 14

21 7.2: HP MP AI AI AI AI 7.2 AI AI AI AI MP MP Sigmoid(x) = Exp( (x 8) (7.1) 7.1 w j 15

22 7.1: AI D A B EuT ables(u, D) (a n, b m ) EuT able(u, s i ) [9] (a) s i (a n, b m ) (b) 16

23 7.2: (c) 2. EuT able(u, s i ) s i a i p(s i, u, a i, EuT able) (a) AI 17

24 (a 1, b 1 ) (a 1, b 2 ) (a 2, b 1 ) (a 2, b 2 ) a 2 7.3: b 1 b 2 a a (a 2, b 1 ) (a 1, b 2 ) b 2 (a 2, b 2 ) 0.45 (a 1, b 2 ) (a 2, b 1 ) (a 1, b 1 ) a 1 A cut ( A) a c ( A cut ) 7.4: b 1 b 2 a a

25 (b) 7.2 Em(a c, s i ) = max c ((1 belief) (rist c average c )) (7.2) max c a c average c a c risk c a c belief AI AI belief 0.5 belief (c) s i a i 7.4 Em 1 e sumem = c Exp(Em(a c, s i )) (7.3) Exp(Em(a i, s i ) ( a p(a i, s i ) = i A cut ) sumem 0 (otherwise) (7.4) 3. ϵ u t 7.1 u t maxabs 7.5 w i w i = w i maxabs (7.5) 19

26 u u D u t u t u s t u t AI u t EuT able(u t, s t ) EuT able(u t, s t ) 7.3 AI AI 1. AI MCAI AI 1 AI (a) (b) (c) MP MP 20

27 0 2. AI AI AI 3. AI AI HP AI 4. AI AI AI 7.4 AI UCB AI AI AI 7.5 AI

28 AI(a) AI(b) AI : MCAI(a) MCAI(b) MCAI(c) MP : MCAI(a) MCAI(b) MCAI(c) MP AI AI 1. AI 22

29 AI 7.7 a 1 a 2 b 1 b 4 AI AI b 1 b 3 b 1 (a 1, b 1 ) b 3 (a 2, b 3 ) b 4 (a 2, b 4 ) b 2 b 1 AI 7.7 b 2 b 4 AI 7.7: b 1 b 2 b 3 b 4 a a AI AI AI AI AI AI AI 23

30 7.8: MCAI(a) MCAI(b) MCAI(c) MCAI(a) 0.000% 6.236% % MCAI(b) 1.930% 0.303% % MCAI(c) 1.589% % 0.213% AI 0.304% 4.175% 4.868% 24

31 8 AI RPG AI 25

32 AI 9.1: 2 MCAI(a) HP MP MCAI : 2 MCAI(b) HP MP MCAI % MCAI(a) 26

33 9.3: 2 MCAI(c) HP MP MCAI % MCAI(b) MCAI(c) MP

34 3. (a) (b) (c) (a) (b) 4. 2 AI a a a a a 1 a 2 a 3 a 4 a 2 2 AI a a a a a 2 a 1 a 4 a 3 a 2 a 1 a 2 28

35 b 3 AI 9.4: 2 1 b 1 b 2 b 3 b 4 a a a a : 2 2 b 1 b 2 b 3 b 4 a a a a

36 30

37 [1] Sander Bakkes, Pieter Spronck and Eric Postma TEAM : The Team-oriented Evolutionary Adaptability Mechanism, Entertainment Computing - ICEC 2004, pp , [2] Aswin Thomas Abraham and Kevin McGee, AI for Dynamic Team-mate Adaptation in Games, Computational Intelligence and Games(CIG), 2010 IEEE Symposium on, pp , [3] Kenneth O.Stanley, Bobby D. Bryant, Student Member, IEEE, and Risto Mikkulainen, Real-Time Neuroevolution int the NERO Video Game, Evolutionary Computation, IEEE Transactions on, Vlo.9, No.6, pp , [4],,,, pp.1-28, [5],, pp [6], The 17th Game Programming Workshop 2012, IPSJ Symposium Series Vol.2012, No.6, pp , [7] Remi Coulom Computing Elo Ratings of Move Patterns in the Game of Go ICGA journal, Vol.30, No.4, [8], VS, [9] Gelly, G., Wang, W., Munos, R. and Teytaud, O. Modification of UCT with Patterns in Monte-Carlo Go Technical Report, No.6062, INRIA,

Copyright 2008 by Tomoyoshi Yamazaki

Copyright 2008 by Tomoyoshi Yamazaki JAIST Reposi https://dspace.j Title 工程管理から知識創造へ - クリニカルパスによる医療のナレッジマネジメント - Author(s) 山﨑, 友義 Citation Issue Date 2008-03 Type Thesis or Dissertation Text version author URL http://hdl.handle.net/10119/4260