Information Technology Reference
In-Depth Information
During a soccer match a coach might have several responsibilities which include i)
monitoring teams performance during a match; ii) advising team players to employ
appropriate tactics by reusing knowledge from previous games; iii) detect high-level
events (e.g. ball possession time) from observing the match observation; iv) classify-
ing opponent strategies (e.g. recognize formations); and v) discovering behavioral
patterns of opponent's strengths and weaknesses (e.g. opponent team neglects the right
side of their defense when defending). Concerning the advice, players as autonomous
agents might decide to follow coach advices or not. On the strategy issue, the sooner
the opponent's team strategy is recognized, the sooner the coach can advise his play-
ers of the best counter-strategy in order to have a higher impact.
Compared to players, a coach normally has more a priori knowledge, a better view
of world and more computational resources available. The communication from
coach-to-players during a match can be achieved using structured coaching languages.
Several works have focused on using a coach to improve the performance of a
team which included i) changing the team formation based on the score difference,
the match remaining time and the ball's path [64]; ii) generating adequate counter
strategies based on the modeling of opponent behaviors using classification models
based on decision trees [65, 68, 70], neural networks [66], naive Bayes [67] and case-
based reasoning [69]; iii) building a marking table that assigns preliminary opponents
to each teammate [71]; and iv) recognizing opponents players physical abilities [71].
3.5
Setplays
Setplays can be understood, in a broad sense, as multi-agent plans that need the com-
mitment of several players in order to reach a common goal. Setplays are very com-
mon in most sports, e.g., soccer, rugby and handball, which can make one believe that
such constructs can also play a useful role in robotic soccer.
The concept of Setplay in RoboCup was first presented in a teamwork and com-
munication strategy for the 2D Simulation league, by Stone and Veloso [50]. These
Setplays, however, were quite limited and were meant to be used only in very specific
situations, like corner kicks and throw-ins, which are decided by the referee, and are
unique for each of these situations.
An interesting approach is presented in Castelpietra et al.[72], where Setplays are
represented as transition graphs. These plans, which are formally defined, have a high
level of abstraction, and can be applied to different robotic platforms, as it has been
the case with Middle-size and four-legged robots. The actual execution of plans and
how the robots deal with synchronization issues are unclear topics. In a related re-
search effort [73], Petri Nets have been used to structure the development of a joint
team with robots from two distinct institutions.
Kok et al. [52, 53] and Kok and Vlassis [74] in their Coordination Graphs (CG),
exploit the dependencies between agents and decomposing a global payoff function
into a sum of local terms that are communicated among agents. Nodes in the CG
represent agents and its edges define dependencies between them, which have to be
coordinated. The continuous aspect of state spaces in robotic soccer discourages the
direct application of CGs. To solve this question, roles are allocated to agents to dis-
cretize this space and then the CG methods are applied to the derived roles. To simpli-
fy the algorithm, it is assumed that only a limited number of near players need to
coordinate their actions.
