Proposta de algoritmo de tomada de decisão para programação de robôs participantes da RoboCup

Abstract

Aiming to promote research on fields such as artificial intelligence and robotics, RoboCup promotes annual robot soccer competitions. These competitions demand work on different knowledge fields like robotics, artificial intelligence, computer vision, electronics, etc. In 2015, the Taura Bots, robotics football team of Federal University of Santa Maria, participated for the first time in the RoboCup competition. Thus, came up the need to enhance the motor skills and the behaviour responsible for controlling the robot. In this context, was create a simulator called TauraSim, to assist on creation of the behaviour for the robot, abandon the need to have the robot always available. During a football match, there are many situations where the agent have more than one option of action, which can be a pass, kick, dribble, etc. With this in mind, it becomes necessary the creation of an algorithm that can be able to execute the best action given the situation of the match. Using the TauraSim simulator it was possible to create a decision-making algorithm for a specific situation of the match, making use of the Naive Bayes classifier to calculate the probabilities, getting considerable results, with a high percentage of correct answers. Another positive aspect is the fact that the Naive Bayes algorithm has its training data apart from its main code, it means that the algorithm can be easily adjusted to other match situations, just by modifying its training data.