Information Technology Reference
In-Depth Information
behaviors (e.g. marking) involve positioning decisions (e.g. move to intercept the
ball). Defensive positioning is an essential aspect of the game, as players without the
ball will spend most of their time moving somewhere rather than trying to intercept it.
Initial strategies for choosing an opponent to mark were essentially reactive and
consisted on each teammate marking the closest opponent but soon became more
elaborate. In 1999, Stone and Veloso [79] proposed that a player who had been as-
signed the role of team captain as part of a LRA, executes a preset algorithm to decide
and communicate to his teammates which opponent they should mark. In 2006, Stol-
zenburg et al. [71] proposed the use of a centralized and decentralized matching algo-
rithms, to determine which opponent a teammate should mark. These algorithms are
executed during non play-on modes and use a distance-based ranking for calculating
the matches. A set of teammates and a set of opponents are selected based on their
relevance for the current situation and their player types. The centralized approach is
executed by a coach which calculates a minimal matching between these sets and
informs all players of his results using communication. The decentralized approach is
executed by each player who tries to find a partial matching if it exists, but it is sus-
ceptible to inconsistencies due to the environment's partial observability which might
lead to inaccurate perceptions of the world.
In 2007, Kyrylov and Hou [80] defined collaborative defensive positioning as a
multi-criteria assignment problem with the following constraints: i) a set of n defend-
ers are assigned to a set of m attackers; ii) each defender must mark at most one at-
tacker; and iii) each attacker must be marked by no more than one defender. In 2010,
they applied the Pareto Optimality principle to improve the usefulness of the assign-
ments by simultaneously minimizing the required time to execute an action and the
threat prevented by taking care of an attacker [81]. Threats are considered preemptive
over time and are prevented using a heuristic-criterion that considers: i) angular size
of own goal from the opponent's location; ii) distance from the opponent's location to
own goal; and iii) distance between the ball and the opponent's. This technique
achieves good performances while gracefully balance the costs and rewards involved
in defensive positioning, but it does not seem to deal adequately with uneven defen-
sive situations such as outnumbered defenders and/or attackers.
Choosing the opponent to mark based only on its proximity might not always be
suitable as it disregards relevant information (e.g. teammates nearby) and will lead to
poor decisions. Also, the use of a fixed centralized mediator (e.g. coach) to assign
opponents to teammates although faster to compute has a negative impact in players
autonomy. With the exception of non play-on periods, this approach is not robust
enough due to the communication constraints of the robotic soccer domain and be-
cause it provides a single point of failure. In 2009, Gabel et al. [82] proposed the use
of a NeuroHassle policy to train a neural network with a back-propagation variant
of the Resilient Propagation Reinforcement Learning (RPROP-RL) technique in
order for a player to learn an aggressive marking behavior which would influence its
positioning.
Search WWH ::




Custom Search