On the Progress of Soccer Simulation Leagues

On the Progress of Soccer Simulation Leagues

On the Progress of Soccer Simulation Leagues Hidehisa Akiyama1 , Klaus Dorer2 , Nuno Lau3 1 2 Fukuoka University, Japan, [email protected] Hochsc...

2MB Sizes 0 Downloads 8 Views

Recommend Documents

soccer leagues - Marquette University Law School
Principal Owner: Clark Hunt, Ron Pizzuti, and a group of local investors ... Facility Financing: The Hunt Sports Group p

On the Epistemology of Computer Simulation
the sense of »media theory« (Medientheorie) it is already to be found in almost every imaginable field of ... even bro

Soccer on - RCN
Marseille, and Paris Saint Germain. Russian Premier ... Teams to watch out for: Zenit St. Petersburg,. Rubin Kazan, CSKA

The simulation of sums
An alternatlve model was proposed by Reeves (1Q84). He also works wlth thls llmltlng dlstrlbutlon, but departs from the

Research on the Numerical Simulation of Sleeper in the Pipeline
Research on the Numerical Simulation of Sleeper in the Pipeline Global. Buckling Controlling Practice. Wen-Bin LIUa, Ai-

Progress of the College
with special reference to Rani Lakshmi Bai and Begum Hazrat Mahal”. •. December 20-22, 2013 – Archival Exhibition

Update on the Progress of the Texas High Performance Schools
May 6, 2012 - Timeline of Consortium Activities to Date. 6. Positive Results of .... cipate in the Consortium, one distr

Update on the Progress of the State Advisory Council Webinar
Jan 23, 2013 - Zelda. VDSS [email protected] 801 East Main. Street. Richmond. VA .... DC Mayor's Office charl

ABOUT The NegrO leAgUes - Scholastic
Before Jackie Robinson joined the Brooklyn. Dodgers in 1947, breaking baseball's “color barrier,” he had played for

Persian Gulf Soccer 2D Simulation Team Description - RoboCup 2017
Abstract. In this paper, algorithms, methods and goals in Persian Gulf. Soccer 2D Simulation team has been described. Th

On the Progress of Soccer Simulation Leagues Hidehisa Akiyama1 , Klaus Dorer2 , Nuno Lau3 1

2

Fukuoka University, Japan, [email protected] Hochschule Offenburg, Germany, [email protected] 3 DETI/IEETA, Aveiro University, Portugal, [email protected]

Abstract. Soccer simulation league is one of the founding leagues of RoboCup. In this paper we discuss the past, present and planned future achievements and changes. Also we summarize the connections and inter-league achievements of this league and provide an overview of the community contributions that made this league successful.

1

Introduction

The soccer simulation league is one of the founding leagues of RoboCup [11]. From the first year of RoboCup it was played with teams of 11 versus 11 players. It is therefore the league that sets the standards for collaboration, team play and opponent modelling. It is also the league in which learning algorithms play a key role since the simulated robots are not subject to wear and tear. After many discussions, in 2004 the league was split into the 2D and 3D Simulation branch. The goal has been to still have a league with 11 player teams, but add another league that keeps the strengths of simulation leagues, yet closes the gap to real robot leagues. The rest of the paper is organized as follows: section 2 describes the development and future of 2D soccer simulation, section 3 the development and future of 3D soccer simulation. In section 4 we point out the achievements and connections to other RoboCup soccer leagues and real soccer. Section 5 summarizes the wealth of contributions that made this league possible.

2

2D Simulation

2.1

History

The RoboCup Soccer Simulator (RCSoccerSim) is the official simulator for the 2D Simulation league since the first RoboCup competition in 1997. RCSoccerSim was designed and developed by Itsuki Noda[19] and later developed and maintained by the RoboCup Soccer Simulator Maintenance Committee as an Open Source project. All software for 2D Simulation is freely available from the official site4 . 4

http://sf.net/projects/sserver

Fig. 1. A screenshot of 2D soccer simulator.

The 2D Simulation league is designed as a soccer competition which is played on a virtual 2D plane field (Fig.1). All objects, such as the ball and players, are modeled as circle on a 2D plane, therefore players never jump and kick the ball into the air. The players’ actuators are also simplified and very much different from real robots. However, almost all soccer rules are implemented and the simulator provides a completely distributed multiagent system which realizes full 11 vs 11 soccer games. The aim of these simplifications is to encourage RoboCup teams to concentrate on the research of teamwork. With progress in a gameplay level, various rule changes have been made to push the progress by the league: 1998 Two important features were implemented. 1) A goal keeper was introduced. This special player possesses an additional action namely ’catch ball’. 2) An offside rule was implemented in the automatic referee. By introducing these features, the human soccer rules are almost fully implemented. Moreover, a stamina model was introduced. It imposes a resource model for the players. To perform actions, stamina needs to be invested and the model implements short-term expenditures with long-term recovery phases. This encourages the development of strategic resource management. 1999 The turn neck command was introduced. This command enables players to change their head direction. Because players’ visible area is restricted, this command enables players to gather more environment information without interfering with body actions. In order to encourage teams to consider online adaptation, an online coach was introduced. Each team can run one online coach agent. The online coach can observe the whole field during a game and can send advice messages to players. 2000 The the ball kick power was increased. This change accelerated the game pace. As a result, the competitions became more attractive. 2001 Heterogeneous players and player substitution were introduced. Teams still can use the default (homogeneous) player, but can select a heteroge-

neous player if necessary. Only the coach can substitute players during nonplay on period. Because the physical abilities of heterogeneous player are randomly generated for each game, online coaches have to deliberate the player assignment according to their team strategy. Also a new contest was added: the Coach Competition. In this competition the coach agent has the ability to advise a team trying to improve its performance. To allow for universality of communication, a standardized coach language was introduced. The best coach was selected as the winner. The rules of this competition have changed frequently and in 2005 the task of the coach became that of identifying strategies (opponent modelling) instead of providing efficient strategies for the own team. 2002 In order to introduce new challenges related to the communication, the length of auditory communication messages among players was shortened. Compensating the bandwidth restriction, two new commands, attentionto and pointto, were introduced. The attentionto command can be used to focus players’ attention on a particular player’s auditory message. If attentionto is off, the player will hear one auditory message from each team selected randomly. The pointto command enables players to point to a spot on the field by their virtual arm and other players can observe the arm direction. This means players can send a location, and indirectly their intention to other players as a visual message. As the improvements of the dribbling skill progressed further, it became too difficult for defenders to block the smart dribbler. The new tackle command enables players to kick the ball in a range wider than the kick command succeeding with a probability based on the ball position relative to the tackling player yet causes immobility for a few cycles. 2003 An automatic penalty mode was added to the simulator. If the game ends in a draw and the simulator option is enabled, the automatic penalty mode is started. As a RoboCup-based testbed, a simplified scenario was implemented, the so-called Keepaway problem, as a testbest for machine learning In this game, the left team (the ”keepers”) attempt to keep the ball away from the right team (the ”takers”). Details of the Keepaway task as well as some experiments can be found in [26]. 2008 The number of heterogeneous player types was increased from 7 to 14. Moreover, assigning the default player type was forbidden except for the goal keeper, and assigning one heterogeneous player type to several players was also forbidden. This means a team has to use at least 10 heterogenous player types. The role assignment problem thus became more difficult. The effective range of the goal keeper’s catch action was reduced. This change was introdeced to encourage teams to consider more effective positioning, not only for the goal keeper, but also for the defensive players. 2009 The stamina capacity model was introduced. This model restricts the total amount of stamina value for each player during the game. Players have to consider their stamina management more carefully and online coaches have to consider the timing of player substitution according to the game situation and the players’ tiredness. The dash model was extended to enable players

to accelerate their body to 4 directions. Players can adjust their position more flexibly and quickly by using the extended model. But, it requires more complicated planning for all move actions. 2010 The tackle command was extended to introduce an intentional foul action. When the intentional foul action is performed, the success probability of action is increased but the player may be penalized. The foul detect probability is defined for each heterogeneous player type. If an intentional and dangerous foul is detected by the automatic referee, the referee penalizes the player by giving the yellow or red card. A new heterogeneous parameter that affects the goal keeper’s catch range was introduced. Teams are still allowed to assign the default type player to the goal keeper in order to keep the compatiblity with previous team binaries. The dash model parameter was changed. Players were now able to accelerate their body to 8 directions. 2.2

Present and Future

Although the community held numerous discussions for introducing new features, no major changes were introduced since 2010. This does not mean the 2D Simulation league stops its progress. Gabel et al. evaluated the performance improvement of teams quantitatively [8]. They compared the performance of teams that participated in the competitions from 2003 to 2007 which has been possible, because no major changes were applied in that period. The results showed that newer champion teams would always overcome older champions. It is necessary to fix the simulator’s specification for some years. Otherwise, researchers have to consume much time to adapt their team to new rules. Because the 2D simulator is already a matured environment, new features will be introduced gradually. Of course, new challenges that discover new research are always required. Although an online coach is available since 1999, online game analysis and online adaptation are still important research topics in the 2D simulation league. In the past few years, several teams started to prepare more than one strategy and switch them according to the opponent team. The role of online coach must become more important in next few years. With the improvement of online coach, the standard coach language will also be updated. Another important topic is the collaboration with real human soccer. Because the 2D Simulation league focuses on the research of teamwork, 2D Simulation has begun to apply decision making techniques and game analysis techniques not only to 2D league but also to human soccer like in [1].

3

3D Simulation

In 2004 the 3D soccer simulation competition was born. Its main goals are those of the 2D soccer simulation, i.e., to keep the focus on multiagent system coordination research and to use the new 3D simulator to conduct research that cannot be performed using real robots, either because of time, money or hardware constraints. It adds the third dimension to the game seeking to make it more realistic and uses more realistic robot models and environment dynamics.

3.1

History

2004 The first RoboCup 3D Simulation Competition took place in 2004. This competition used the Simspark generical physical simulator platform [20] to build a 11 against 11 soccer game simulator called rcssserver3d. The simulation includes several new innovations such as (obviously) the 3D model of the environment, but also the use of the ODE physics engine library to model and update the dynamics of simulated objects and a middleware for agent simulations called SPADES [23] that manages the distributed simulation, and ensures results do not depend on network or system load, and a new timing model, where agent thinking times are taken into account. To keep the introduction of the 3D environment in line with 11 vs 11 games, the 3D robot model used in 2004 was quite simple and its shape was a sphere, as can be seen in Fig. 2. All actions were performed by applying forces either to the agent or to the ball. 2005 The 2005 rules and models were essentially identical to the ones used in 2004. However, there were several changes inside the simulator to make it more efficient and to remove bugs. Big effort was also spent in the documentation of the new simulator as in its initial year, the documentation was quite scarce. 2006 The robot vision perceptor was changed in 2006 to restrict agents vision to a limited field of view. To control the looking direction a new PanTilt effector was added to the robot. A limited bandwidth broadcast communication model was also introduced in this year. The soccer rules were also changed to approximate FIFA rules, and an offside rule, identical to 2D rule, was added to the simulator. Optimizations made to the simulator turned it into a much more efficient application. 2007 In the 2007 competition, the robot model was changed from the simple sphere model towards the targeted humanoid robot model. The humanoid model used in 2007 was based on real humanoid HOAP2 from Fujitsu [3], which can be seen in Figure 2. Each of the joints could be controlled setting the desired angular velocity. The new model introduced many new research challenges to the Simulation League, enabling this league to conduct research on humanoid robotics. The trade-off was that the number of players per team had to be reduced from 11 to 2, as the simulator could not cope with more than 4 humanoid agents in total, and the restricted vision was also changed to a global vision perceptor. The development of efficient humanoid skills became the most important point for the success of the team, making coordination research less important in this year. Nevertheless, this change was necessary to bring Simulation League research closer to the real robots and, as predicted, the number of players per team was increased in the following years making coordination essential again. 2008 With the introduction of Aldebaran’s Nao Robot in the Standard Platform League, also 3D soccer simulation switched to the simulation of Nao robots. In 2008 games were played 3 versus 3. With this step, the 3D soccer simulation league again made a step towards narrowing the gap between soccer simulation leagues and real robot leagues.

Fig. 2. Robot models used. Spheres (2003-2006), Hoap robot model [3] (2007) and NAO (from 2008).

2009 2009 saw an increase of the number of robots per team to 4 with the goal in mind to reach 11 vs 11 players eventually. The robots’ omnidirectional view was replaced by a restricted view with a cone similar to real Nao robots. 2010 The major change for 2010 in Singapore has been the increase of teams of 6 simulated Nao players. Better hardware and software optimizations made it possible to increase the number of players and the field size accordingly. Also, to boost the development of simulation server and infrastructure, a 3D development competition was held. 2011 The last intermediate step to 11 versus 11 was taken in 2011 with teams of 9 versus 9 NAOs. Again the field size has been increased to keep the space per robot roughly the same. The visualization was done using a new visualizer RoboViz [25]. With a special 3D beamer and glasses the games could be watched in 3D in Istanbul. 2012 A major milestone has been achieved in 2012 simulating games of 11 vs 11 NAO robots for the first time. The soccer field size has been increased to represent the size of a real soccer field with respect to the size of the robots. As of 2012, both soccer simulation leagues were the only leagues with the full final team size played in 2050. 2013 A major strength of simulation leagues was first exploited in 2013 with the introduction of heterogeneous robots. In no other league it is so cheap to change robot models and use different variations of a robot model in one and the same game. The rational behind introduction variations of the NAO robot has been to shift from programming good behavior skills for a very specific robot to creating good algorithms for the behaviors to any similar robot, especially for walking. The exact robot models have only been published some days before the competition. Only teams that are able to adjust their algorithms and have the modified robots learning the skills succeed. Also a drop-in player challenge was held in conjunction with several other leagues. Mixed teams with two players from each participant team played against other mixed teams put together from other participating teams. The challenge is to play with other players without a common strategy agreed upon. A couple of games with a changed mix of teams determined the winner.

Fig. 3. 11 vs 11 Nao robot game.

3.2

Present and Future

For 2014, a new goal has been introduced in 3D soccer simulation: the league committed itself to work towards having the first running robots. In a running robots challenge, robots are evaluated for their run speed, but also on how much time both feet are off the ground. This is possible in 3D soccer simulation since the simulated Nao robots have slightly stronger motors than real Nao robots. Nao models with toes were introduced for this challenge to allow more humanlike walking and running. The idea behind is to explore in a cheap way what hardware setups are required to make running with biped robots possible. Also, while in 2013 the usage of heterogeneous robot models was optional, in 2014 teams are forced to use heterogenous robots. At most 7 robots may use the standard NAO type. The remaining 4 robots have to use 3 differnt variations of the NAO. The main challenge for the future will remain to find a good balance between having a multiagent simulation for intelligent robotics research with ideally 11 vs 11 games and having a more realistic simulation in terms of real hardware. Having the 11 vs 11 goal achieved, concrete suggestions have been made for the later: a model for energy consumption and motor warming has been implemented and suggested, stiffness of motors, or more realistic noise models for sensors. Also a transition to new, more realistic simulators like gazebo is an option. Above that, 3D Simulation should also play a role in research for new robot models and hardware. This has started with the introduction of heterogeneous robot models including a model with toes. It is now continued with the running challenge in which teams for the first time can suggest their own robot model variations. And it should be further developped in the future to, for example, use new sensors and actuators if these sensors are biologically plausible and likely to be developed on real robots. Suggestions include linear actuators, touch suites and many more.

4

Inter-League Achievements

Technologies that result from research in the Simulation League are often used in other RoboCup leagues. In some cases, the same institution creates new teams to compete in new leagues using the knowledge from a previous Simulation League team or starts its participation at several leagues simultaneously, this is the case, for example, of CMUnited (2D, Small, MSL, 4legged, SPL), Brainstormers (2D, MSL, 3D), FC Portugal (2D, 3D, Mixed Reality, 4legged, SPL, MSL), UT Austin Villa (2D, 3D, SPL), WrightEagle (2D, 3D, 4legged,SPL), Bold Hearts (3D, Humanoid) or magmaOffenburg (3D, Humanoid). In other cases the technologies are adopted by completely independent teams from different leagues. One of the major challenges in soccer is the positioning of the team in the field. Several positioning systems have been proposed in the Simulation League [27, 21, 4, 2, 14] and have been used in other leagues [27, 12, 10]. Simulation is a very adequate environment for the automatic development of behaviors. Several teams have developed machine learning to enhance their 2D and 3D teams. A very significant example is the research developed by the Brainstormers team, that has been applied in the Simulation League and also in the Middle-Size League [22, 7]. Using humanoid models at the 3D Simulation league fostered automatic humanoid behavior generation research. The techniques that have been developed in the simulation league are mostly based on optimization and machine learning, either using model-free or model-based approaches and have had a strong impact on real robot leagues [5, 6]. It is very interesting to see that, as referred to before, several 3D Simulation Teams have created teams that participate in the Standard Platform League or at the Humanoid League. Teams that maintain a strong link to Simulation League research and developments have been very successful in other leagues. There are several examples of this kind of teams that became RoboCup champions in SSL (CMUnited, CMDragons), 4Legged (CMPack), MSL (Brainstormers Tribots, CAMBADA) and SPL (UT Austin Villa).

5

Community Contributions

More than in most leagues, soccer simulation leagues’ progress is heavily influenced by community contributions. The main workhorses for 2D and 3D Simulation are the soccer simulators. Both have been fully developed and maintained by the community specifically to ensure independence from external vendors and so that the league retains the right to fully use and modify them as the requirements of the league progress. Around the simulators, a wealth of other software was created to visualize, analyze and comment games as well as run games or complete tournaments. 5.1

Simulator Development

The original 2D soccer simulator was proposed and developed by Noda et al. in 1997 [19]. Many people contributed to the development under the lead of Itsuki

Noda, Tom Howard and later and until now by Hidehisa Akiyama. The simulator runs on Linux and has reached version 155 . A more detailed history of changes is provided in section 2.1. The 3D Simulator6 was initiated by work from Markus Rollmann, Oliver Obst [20], Jan Murray and Joschka Boedecker who provided the Spark simulator as a generic simulator for 3D. It is based on ODE physics engine7 to simulate physically realistic objects in 3D. The actual soccer simulation is a separate module based on top of Spark. It provides the robot models, soccer rules, automated referee and more. It proved its flexibility many times, for example whenever new robot types were used in the simulation. Today it provides the base for having heterogeneous robots types available in one and the same game. Again many people from the community contributed to the server development under the lead of the aforementioned and later lead by Hedayat Vatankah.

5.2

Visualizers

To the public, the quality of the simulations is mainly associated with the quality of visualizing it. A couple of 2D visualizers are available for free for Linux and Windows, many of them adding debugging features and logfile replaying. 3D Simulation offers two visualizers, rcssmonitor3D and RoboViz [25]8 . The later is

Fig. 4. Debugging features of RoboViz [25].

now used as main visualizer in competitions and by many teams for team-specific graphical visualization of real time debugging information. 5 6 7 8

http://sourceforge.net/projects/sserver/ http://sourceforge.net/projects/simspark/ http://ode.org/ode-latest-userguide.html http://sites.google.com/site/umroboviz/

5.3

Source Code Releases

It is a condition of participation that binaries of all teams are released after a RoboCup. This allows all teams to test against old teams and to perform scientific research including other team’s binaries. Source code release, however, is voluntary. Nevertheless, more than 15 teams have released source code libraries or fully functional agent code in 3D. Many teams in 2D based their code on source code bases from CMUnited 99, FCPortugal, UVATrilearn and Helios. The code is available for C++, Java, C#, Clojure, C, Prolog and Javascript. This way, new teams do not have to redevelop low level communication or geometrical transformations, but can focus on high level skills and decision making 9 .

5.4

Benchmarks

Keepaway (see section 2.1), a subtask of 2D Simulation has been a well established benchmark domain in the reinforcement community for many years [26]. The magmaRunChallenge is a currently developed benchmark tool for the 3D running robot challenge in which robots are benchmarked for speed and the relative amount of time both feet are off the ground.

5.5

Miscellaneous

Many more contributions added to the success of soccer simulation league. To only mention a few: the league managers, a comprehensive set of scripts to run tournament rounds in a completely automated way, a communication proxy that decouples the simulation from any network or simulation server performance issues in 3D, a framework to host a 2D tournament on a remote system, a couple of live commentary systems [18, 28, 9] or the ssil, an automated internet league.

6

Conclusion

In this paper we have summarized the past, present and future of the 2D and 3D RoboCup Soccer Simulation leagues as an excellent domain for multi-agent, machine learning and humanoid robot research. Due to its low cost, it is a good entry point for new teams into the RoboCup community. Many teams with roots in soccer simulation now participate in hardware leagues with some of them even winning other league competitions. The success of the soccer simulation league is due to the inpour of community contributions and the perseverance of the simulator maintainers which is an essential tool for mapping the path towards the Grand Challenge of 2050. 9

http://wiki.robocup.org/wiki/Soccer Simulation League

References 1. Abreu, P., Moreira, J., Costa, I., Castelo, D., Reis, L., Garganta, J.: Human versus virtual robotics soccer: A technical analysis. European Journal of Sport Science 12(1), 26–35 (2012) 2. Akiyama, H., Noda, I.: Multi-agent positioning mechanism in the dynamic environment. In: Visser, U., Ribeiro, F., Ohashi, T., Dellaert, F. (eds.) RoboCup 2007: Robot Soccer World Cup XI, Lecture Notes in Computer Science, vol. 5001, pp. 377–384. Springer Berlin Heidelberg (2008) 3. Boedecker, J., Dorer, K., Rollmann, M., Xu, Y., Xue, F., Buchta, M., Vatankhah, H.: Simspark users manual. Version 1, 17–18 (2008) 4. Dashti, H., Aghaeepour, N., Asadi, S., Bastani, M., Delafkar, Z., Disfani, F., Ghaderi, S., Kamali, S., Pashami, S., Siahpirani, A.: Dynamic positioning based on voronoi cells (dpvc). In: Bredenfeld, A., Jacoff, A., Noda, I., Takahashi, Y. (eds.) RoboCup 2005: Robot Soccer World Cup IX, Lecture Notes in Computer Science, vol. 4020, pp. 219–229. Springer Berlin Heidelberg (2006) 5. Domingues, E., Lau, N., Pimentel, B., Shafii, N., Reis, L.P., Neves, A.J.: Humanoid behaviors: from simulation to a real robot. In: Progress in Artificial Intelligence, pp. 352–364. Springer Berlin Heidelberg (2011) 6. Farchy, A., Barrett, S., MacAlpine, P., Stone, P.: Humanoid robots learning to walk faster: From the real world to simulation and back. In: Proc. of 12th Int. Conf. on Autonomous Agents and Multiagent Systems (AAMAS) (May 2013) 7. Gabel, T., Hafner, R., Lange, S., Lauer, M., Riedmiller, M.: Bridging the gap: Learning in the robocup simulation and midsize league. In: Proceedings of the 7th Portuguese Conference on Automatic Control (Controlo 2006). Portuguese Society of Automatic Control, Porto, Portugal (2006) 8. Gabel, T., Riedmiller, M.: On progress in robocup: The simulation league showcase. In: del Solar, J.R., Chown, E., Pl¨ oger, P.G. (eds.) RobuCup. Lecture Notes in Computer Science, vol. 6556, pp. 36–47. Springer (2010) 9. Jung, B., Oesker, M., Hecht, H.: Virtual robocup: Real-time 3d visualization of 2d soccer games. In: Veloso, Pagello, Kitano (eds.) RoboCup-99: Robot Soccer World Cup III (LNAI 1856). Springer-Verlag (2000) 10. Kaden, S., Mellmann, H., Scheunemann, M., Burkhard, H.D.: Voronoi based strategic positioning for robot soccer. In: Proceedings of the 22nd International Workshop on Concurrency, Specification and Programming (CS&P). vol. 1032, pp. 271– 282. CEUR-WS.org (2013) 11. Kitano, H., Asada, M., Kuniyoshi, Y., Noda, I., Osawa, E.: Robocup: The robot world cup initiative. In: IJCAI-95 Workshop on Entertainment and AI/Alife. Montreal, Quebec (August 1995) 12. Lau, N., Lopes, L., Corrente, G., Filipe, N., Sequeira, R.: Robot team coordination using dynamic role and positioning assignment and role based setplays. Mechatronics 21(2), 445–454 (2011), cited By (since 1996)6 13. Lau, N., Reis, L.P.: FC Portugal - high-level coordination methodologies in soccer robotics. In: Lima, P. (ed.) Robotic Soccer. I-Tech Education and Publishing, Vienna, Austria (2007) 14. MacAlpine, P., Barrett, S., Urieli, D., Vu, V., Stone, P.: Design and optimization of an omnidirectional humanoid walk: A winning approach at the RoboCup 2011 3D simulation competition. In: Proceedings of the Twenty-Sixth AAAI Conference on Artificial Intelligence (AAAI) (July 2012)

15. MacAlpine, P., Collins, N., Lopez-Mobilia, A., Stone, P.: UT Austin Villa: RoboCup 2012 3D simulation league champion. In: Chen, X., Stone, P., Sucar, L.E., der Zant, T.V. (eds.) RoboCup-2012: Robot Soccer World Cup XVI. Lecture Notes in Artificial Intelligence, Springer Verlag, Berlin (2013) 16. MacAlpine, P., Urieli, D., Barrett, S., Kalyanakrishnan, S., Barrera, F., LopezMobilia, A., S ¸ tiurc˘ a, N., Vu, V., Stone, P.: UT Austin Villa 2011: A champion agent in the RoboCup 3D soccer simulation competition. In: Proc. of 11th Int. Conf. on Autonomous Agents and Multiagent Systems (AAMAS) (June 2012) 17. Marques, H., Lau, N., Reis, L.P.: Architecture and basic skills of the FC Portugal 3D simulation team. Revista do Departamento de Electrnica e Telecomunicaes da Universidade de Aveiro 4(4), 478–485 (Maio 2005) 18. Matsubara, H., Frank, I., Tanaka-Ishii, K., Noda, I., Nakashima, H., Hashida, K.: automatic soccer commentary and robocup. In: Asadan, M., Kitano, H. (eds.) RoboCup’98: Robot Soccer World Cup II. No. 1604 in Lecture Notes in Artificial Intelligence, Springer Verlag (1999) 19. Noda, I., Matsubara, H., Hiraki, K., Frank, I.: Soccer server: A tool for research on multiagent systems. Applied Artificial Intelligence 12(2-3), 233–250 (1998) 20. Obst, O., Rollmann, M.: Spark - a generic simulator for physical multi-agent simulations. In: Lindemann, G., Denzinger, J., Timm, I.J., Unland, R. (eds.) MATES. Lecture Notes in Computer Science, vol. 3187, pp. 243–257. Springer (2004) 21. Reis, L., Lau, N., Oliveira, E.: Situation based strategic positioning for coordinating a team of homogeneous agents. In: Balancing Reactivity and Social Deliberation in Multi-Agent Systems, Lecture Notes in Computer Science, vol. 2103, pp. 175–197. Springer Berlin Heidelberg (2001) 22. Riedmiller, M., Gabel, T., Hafner, R., Lange, S.: Reinforcement learning for robot soccer. Autonomous Robots 27(1), 55–74 (2009) 23. Riley, P., Riley, G.: SPADES — a distributed agent simulation environment with software-in-the-loop execution. In: Chick, S., S´ anchez, P.J., Ferrin, D., Morrice, D.J. (eds.) Winter Simulation Conference Proceedings. vol. 1, pp. 817–825 (2003) 24. Solgi, M., Dezfouli, S., Baghi, H., Ghaderi, A., Mola, O., Kazempour, V., Akhondian, S., Montazeri, H., Nickabadi, A., Moradi, S.: Aria 2005 3d soccer simulation team description. In: Proc. RoboCup Symposium 2005. Osaka, Japan (2005) 25. Stoecker, J., Visser, U.: Roboviz: Programmable visualization for simulated soccer. In: Roefer, T., Mayer, N., Savage, J., Saranl, U. (eds.) RoboCup 2011: Robot Soccer World Cup XV, Lecture Notes in Computer Science, vol. 7416, pp. 282– 293. Springer Berlin Heidelberg (2012) 26. Stone, P., Kuhlmann, G., Taylor, M.E., Liu, Y.: Keepaway soccer: From machine learning testbed to benchmark. In: Noda, I., Jacoff, A., Bredenfeld, A., Takahashi, Y. (eds.) RoboCup-2005: Robot Soccer World Cup IX. Springer Verlag, Berlin (2006) 27. Veloso, M., Stone, P., Bowling, M.: Anticipation as a key for collaboration in a team of agents: A case study in robotic soccer. In: Schenker, P.S., McKee, G.T. (eds.) Proceedings of SPIE Sensor Fusion and Decentralized Control in Robotic Systems II. vol. 3839, pp. 134–143. SPIE, Bellingham, WA (September 1999) 28. Voelz, D., Andre, E., Herzog, G., Rist, T.: Rocco: A robocup soccer commentator system. In: Asadan, M., Kitano, H. (eds.) RoboCup’98: Robot Soccer World Cup II. No. 1604 in Lecture Notes in Artificial Intelligence, Springer Verlag (1999) 29. Yuan, X., Yingzi, T.: Layered omnidirectional walking controller for the humanoid soccer robot. In: Robotics and Biomimetics, 2007. ROBIO 2007. IEEE International Conference on. pp. 757–762 (Dec 2007)

Appendix

Table 1. Table of Champions in 2D and 3D Simulation. Year 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

2D champion AT Humboldt CMUnited CMUnited FC Portugal TsinghuAeolus TsinghuAeolus UvA Trilearn STEP Brainstormers WrightEagle Brainstormers Brainstormers WrightEagle HELIOS WrightEagle HELIOS WrightEagle

Country Germany USA USA Portugal China China Netherlands Russia Germany China Germany Germany China Japan China Japan China

3D champion Aria[24] AriaKavir FC Portugal[13, 17] WrightEagle SEU-RedSun[29] SEU-RedSun Apollo3D UT Austin Villa[14, 16] UT Austin Villa[15] Apollo3D

Country Iran Iran Portugal China China China China USA USA China