Information Technology Reference
In-Depth Information
3
Coordination in Multi-robot Systems
One of the most important tasks for players is to select and initiate an appropriate
(possibly cooperative) behavior (e.g. pass to a teammate, open a pass line) in a given
context, using (or not) knowledge from past experiences in order to help their team to
win. Good coordination techniques can help achieve this goal, although their success
is highly dependent on players individual abilities (low-level skills) to execute
adequate competitive decisions. Also, without having a good accuracy on the percep-
tion of the world in a real-time, partially observable, stochastic and dynamic envi-
ronment such as the one in the RoboCup simulated soccer leagues the coordination of
player actions would be much more difficult. This section presents an overview of the
coordination methodologies applied in RoboCup, in the last years, with emphasis on
the simulated 2D league that is, by its characteristics, the best league to research on
the coordination area in the context of RoboCup.
3.1
Communication
Communication can be used by players to exchange their intentions and beliefs about
the world. The exchange of beliefs helps the players to obtain a more accurate percep-
tion of the world, congruent with their teammates. However, not all beliefs have the
same importance being, in the soccer domain, the ball location considered the most
important [48]. A better knowledge of the world empowers players to execute better
suited behaviors.
In several RoboCup leagues a coach agent is available that does not need to rely on
communication because he has a global and error-free view of the world. However,
typically his communication is highly constrained (low bandwidth and/or significant
delays for transmission and reception),making it impossible to rely on his messages to
keep an accurate perception of the world state.
Initially, on the simulation 2d league, player-to-player communication allowed the
transmission of long messages and players could hear all the messages sent in each cycle.
This led to the development of techniques that were not concerned with the selection of
which relevant information should be transmitted at any given time, since the bandwidth
available was enough to share the most meaningful information about one's world state
with his teammates [49] and communicate useful events/opportunities [1]. The size of
messages was shortly reduced to a minimum and the number of simultaneous messages
heard in each cycle reduced to one. These new constraints required agents to be capable
of cautiously selecting the most pertinent information to send and with their teammates at
each instant. Stone et al. [50] addressed the issue of low-bandwidth with the specification
of a Locker-Room Agreement (LRA) in which team members have a common under-
standing of environment cues that trigger predefined strategies. This enabled them to
coordinate by minimizing or even eliminating the need for communication. Reis and Lau
[1, 2] tried to tackle this issue by measuring the importance of each piece of information
through utility metrics based on the current match situation and on the estimated
knowledge of teammates. This idea afterward extended in [21] with a Situation Based
Communication framework.
Search WWH ::




Custom Search