Deep Neural Network Algorithm to Control a Curved Kicking Mechanism in RoboCup Small Size League

• Francisco A. B. Azevedo ITA
• Guilherme P. C. Leão ITA
• Marcos R. O. A. Maximo ITA

Resumo

Small Size League (SSL) is a robotics soccer league at the Robotics World Cup, RoboCup. In this league, the proposition of a mechanism capable of performing unpredictable kicks and passes inspired works of both the physical mechanism to implement this and the algorithms to extract the most of this mechanism. The present work's contribution is an improvement on a previously algorithm that aims to perform a real-time inversion of the non-linear ordinary differential equation that models the ball's trajectory to find the parameters to hit a target with the curved kick mechanism, by adding new concepts and by using a Deep Neural Network. New strategies are added to improve the efficiency of the algorithm and to make it easier to be integrated with other algorithms on the robot and also make it capable of finding a trajectory that avoids enemies on the field.

Palavras-chave: Deep learning, Conferences, Neural networks, Education, Ordinary differential equations, Real-time systems, Mathematical models, Robot programming, Robot soccer, Deep Learning, Small Size League, RoboCup