Information Technology Reference
In-Depth Information
Team Agent Behavior Architecture (TABA) [75] uses hierarchical task decomposi-
tions to coordinate the behavior of players in the old four-legged league. Collabora-
tion between players is managed through formations including roles, which describe
players' positions. Formations and role choices depend on the team attitude and game
state. A CBR system stores a strategy base. A strategy case is a plan designed to
achieve a particular goal and includes applicability and termination conditions and a
list of formations with roles. Further work by Vega et al. [76] includes a Soccer Strat-
egy Description Symbols (SSDS) graphical notation, an eXtensible Markup Language
(XML) behavior language and a control simulator based on Finite State Machines.
4
Strategic Positioning, Formations and Setplays
The selection of a good position during a match is a challenging task for players due
to the unpredictability of the environment. However, the likelihood of collaboration in
a match is directly related to the adequacy of a player's position. During a match,
typically at most one player will carry the ball at each instant. For this reason, players
will spend the most time without the ball trying to figure out where to move.
In 1999, Stone [49] proposed a Strategic Positioning using Attraction and Repul-
sion (SPAR) in which a player maximizes the distance to other players and minimizes
the distance to the opponent goal, the active teammate and the ball. Although this
approach enabled a player to anticipate the collaborative needs of his teammates but it
did not allow the team to assume suitable shapes (e.g. compact for defending) for
different situations nor the teammates to have different positional behaviors. Reis et
al. [1, 2, 4, 19, 20] proposed a Situation-Based Strategic Positioning (SBSP) method
in 2001 to do just that. This method defines team strategy as a set of player roles (de-
scribing their behavior) and a set of tactics composed of several formations. Each
formation is used in a strategic situation and assigns each player a default spatial posi-
tioning and a role. In 2006, Dashti et al. proposed a dynamic positioning based on
Voronoi Cells [77] that distributes players across the field making use of attraction
vectors to reflect players' tendency towards specific objects based on their roles and
the current match situation. Additionally, it does not require the use of home positions
nor it limits the number of players per role contrarily to SBSP. In 2008, Akyama et al.
proposed a Delaunay Triangulation method [78] inspired by SBSP which divides the
soccer field into triangles based on training data and builds a map from a focal point
(e.g. ball position) to a desirable positioning for each player. Additionally, this me-
thod supports the use of i) constraints to fix topological relations between different
sets of training data to compose more flexible formations; ii) unsupervised learning
methods to cope with large or noisy datasets; and iii) linear interpolation methods
to circumvent unknown inputs. Besides having a good approximation accuracy, is
locally adjustable, fast running, scalable and reproducible.
4.1
Defensive Positioning
The main goal of a defending team (without the ball possession) is to stop the oppo-
nent's team attack and create conditions to launch their own. In general, defensive
