RoboCup 1 2D 3D Figre 1 2 2D 3D 2D 2D 3D 2D 2D Earth Mover s Distance Earth Mover s Distance 3.1 (x y ) p i w pi Figure 3 opuscom Uv

Size: px

Start display at page:

Download "RoboCup 1 2D 3D Figre 1 2 2D 3D 2D 2D 3D 2D 2D Earth Mover s Distance Earth Mover s Distance 3.1 (x y ) p i w pi Figure 3 opuscom Uv"

こおがのえ
6 years ago
Views:

1 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 (5/3) RoboCup Predicting Game Results using Kick Distributions in RoboCup,, Jordan Henrio,, Satoshi Mifune, Tomoharu Nakashima, Jordan Henrio, Hidehisa Akiyama Soichi Ayai Osaka prefecture University, Fukuoka University Abstract Predicting the game result using kick distributions is studied in this paper. Although it is not possible to exactly know a strategy that a team is taking, that strategy might be well represented by how the players in the team kick the ball during games. Passes and dribbles that are made during a game are extracted to form a kick distribution. It is assumed that the kick distribution represents the strategy of a team. A series of computational experiments are conducted in order to examine the performance of the proposed method. 1 RoboCup[1] RoboCup Gabel [2] Abreu [3] 1 RoboCup 2D Earth Mover s Distance (EMD) [4] 4 2 RoboCup RoboCup RoboCup 2050 RoboCup RoboCup 1

RoboCup 1 2D 3D Figre 1 2 2D 3D 2D 2D 3D 2D 2D 3000 6000 1 0.

1 (x y ) p i w pi Figure 3 opuscom UvA Trilearn opuscom Figure 3 35 30 25 20 15 10 5 1: 2D Simulation League

2 RoboCup 1 2D 3D Figre 1 2 2D 3D 2D 2D 3D 2D 2D Earth Mover s Distance Earth Mover s Distance 3.1 (x y ) p i w pi Figure 3 opuscom UvA Trilearn opuscom Figure : 2D Simulation League : kick distribution that is obtained from a game between opuscom and UvA Trilearn 3.2 Earth Mover s Distance 2: 3D Simulation League Earth Movers s Distance(EMD) EMD [5] EMD 2

3 EMD P P = {(p 1, w p1 ),, (p m, w pm )} P m p i w pi Q Q = {(q 1, w q1 ),, (q m, w qm )} EMD 2 p i q j d ij D = [d ij ] p i q j f ij F = [f ij ] (1) F EMD W = d ij f ij (1) f ij 0 (1 i m, 1 j n) (2) n f ij w pi (1 i m) (3) j=1 m f ij w qi (1 j m) (4) i=1 f ij = min( w pi, w qj ) (5) (2) (3) p i (4) q i (5) F P Q EMD d ij fij EMD(P, Q) = m n f ij (6) W EMD 3.3 Step 1 : EMD Step 2 : Step 3 : Step 4 : 2 Step 5 : 1 2 Step UvA Trilearn(2005)[6] BrainStomers(2009)[7] HELIOS(2014)[8] WrightEagle(2014)[9] 4 opuscom(2014) opuscom x y EMD EMD 3

4.2 5 opuscom 2 UvA Trilearn UvA Trilearn 5.3 opuscom opuscom opuscom Figure 5 5.

4 4.2 5 opuscom 2 UvA Trilearn UvA Trilearn 5.3 opuscom opuscom opuscom Figure UvA Trilearn(2005) BrainStomers(2009) HE- LIOS(2014) WrightEagle(2014) 4 opus- COM(2014) UvA Trilearn BrainStomers HELIOS WrightEagle 4 opuscom Table 1 1: Game results with UvA Trilearn, BrainStormers, HELIOS and WrightEagle Win Draw Lose UvA Trilearn BrainStomers WrightEagle HELIOS Figure 4 5: Clustering result for kick distributions of opus- COM Figure 5 U UvA Trilearn B Brain- Stomers W WrightEagle H HELIOS opuscom opuscom : Clustering result for kick distributions of opponents Figure 4 U UvA Trilearn B Brain- Stomers W WrightEagle H HELIOS 1 Figure 6 Figure 6 U UvA Trilearn B Brain- Stomers W WrightEagle H HELIOS opuscom 2 4

5 6: Clustering result for kick distributions of opus- COM and opponents 7: Labels of clusters : opponent team s kick distribution 5.5 UvA Trilearn BrainStomers HE- LIOS WrightEagle 4 opuscom Table 2 2: Game results with UvA Trilearn, BrainStormers, HELIOS and WrightEagle Win Draw Lose UvA Trilearn BrainStomers WrightEagle HELIOS : Labels of clusters : our team s kick distribution opuscom Figure Table 3 3: The number of correct opuscom UvA Trilearn BrainStomers WrightEagle HELIOS : Labels of clusters : both team s kick distribution 5

6 Table 3 opuscom UvA Trilearn 6 EMD [6] Julle R. Kok and Nikos Vlassis, UvA Trilearn2005 Team Description Paper, RoboCup2005, CD-ROM (5 pages), Osaka, Japan(2005). [7] Thomas Gabel, Martin Riedmiller, BrainStormers 2D - Team Description 2009, RoboCup2009, CD- ROM (6 pages), Graz, Austria(2009). [8] Hidehisa Akiyama, Tomoharu Nakashima, Katsuhiro Yamashita, Satoshi Mifune, HELIOS2014 Team Description Paper, RoboCup2014, CD-ROM (6 pages), JoãoPessoa, Brazil(2014). [9] Haochong Zhang, Guanghui Lu, Rongya Chen, Xiao Li and Xiaoping Chen, WrightEagle 2D Soccer Simulation Team Description 2014, RoboCup2014, CD-ROM (6 pages), JoãoPessoa, Brazil(2014). [1] Hiroaki Kitano, Minoru Asada, Yasuo Kuniyoshi, Itsuki Noda, Eiichi Osawa and Hitoshi Matsubara, RoboCup: A Challenge Problem for AI, AIM agazine, Vol.18, No.1, pp.73-85(1997). [2] Thomas Gabel, Martin Riedmiller On Progress in RoboCup: The Simulation League Showcase The 14th RoboCup 2010 Symposium pp Springer Berlin Heidelberg(2010). [3] Pedro Abreu, João Moreira, Israel Costa, Daniel Castelão, Luis Reis, Júlio Garganta, Human Versus Virtual Robotics Soccer: A Technical Analysis, European Journal of Sport Science 12(1), pp.26-35, Taylor & Francis(2011) [4] Y.Rubner, C.Tomasi and L.J.guibas, The earth mover s distance as a metric for image retrieval, International Journal of Computer Vision, 40(2), pp (2000) [5] Earth Mover s Distance (2007). 6

一般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-04 (3/30) RoboCup Controlling Actions with P

一般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-4 (3/) RoboCup Controlling Actions with Pruning in RoboCup Soccer,, Sho Tanaka, Tomoharu Nakashima,