In recent years there has been a great effort concentrated in the research on multi-agent systems. RoboCup is an international initiative advocated to the stimulation of this research, using the soccer game as a standard platform for benchmarking techniques, prove architectures and devise models of interaction among agents in an opposition environment. One of the problems to consider in RoboCup is the implementation of a vision system, which is the main source of information for agents during games. The present work focuses on the implementation of a robust and fault tolerant global vision system for RoboCup Small League soccer teams. It is based on a vision control approach, in which vision processes are guided by necessity of information and knowledge about the environment. The object detection is based on a chromatic approach where chromatic patterns were modeled using a mixture of gaussian functions, trained with a stochastic gradient descent method. The implemented system meets, and in certain cases exceeds, the functionality required to participate in RoboCup and reported in related works.
[1]
Hans G. C. Tråvén,et al.
A neural network approach to statistical pattern classification by 'semiparametric' estimation of probability density functions
,
1991,
IEEE Trans. Neural Networks.
[2]
Hiroaki Kitano,et al.
The RoboCup Physical Agent Challenge: Goals and Protocols for Phase 1
,
1997,
RoboCup.
[3]
Gordon Cheng,et al.
Real-Time Vision Processing for a Soccer Playing Mobile Robot
,
1997,
RoboCup.
[4]
Ronald C. Arkin,et al.
An Behavior-based Robotics
,
1998
.
[5]
Hiroaki Kitano,et al.
RoboCup: A Challenge Problem for AI and Robotics
,
1997,
RoboCup.
[6]
Hiroaki Kitano,et al.
RoboCup: The Robot World Cup Initiative
,
1997,
AGENTS '97.
[7]
D. Rubin,et al.
Maximum likelihood from incomplete data via the EM - algorithm plus discussions on the paper
,
1977
.