Internship Report

The Robot Soccer World Cup (RoboCup) Federation is an international organization which focuses on the promotion of robotics and Artificial Intelligence research, by offering a publicly appealing challenge: build robots that play soccer. The teams: Tech United and CAMBADA compete in the Middle-Sized League (MSL) among other leagues. In the MSL robots of no more than 50 cm in diameter and 80 cm in height, play soccer in teams of 5 robots. During such a game of soccer, the position of the ball is of great importance, most teams use a catadioptric vision system to detect the ball. With this system it is not possible to find the position of the ball in three dimensions, the ball is therefore projected on the field. Now, through numerous developments in the MSL league, the ball is airborne a great part of the game, increasing the need for 3D ball positioning. One of the methods to obtain this 3D ball position is by triangulating omni-vision camera data of pairs of robots. The research presented in this report is therefore described by the following target: Research and Verify whether cooperative sensing can be used for real-time 3D ball positioning using triangulation of omni-vision camera data. In the process of implementing the triangulation algorithm an existing trajectory generator is improved. Using this trajectory generator a source of inaccuracy is identified in a previously proposed method. The inaccuracy results from extrapolation, which is used to cope with the effects of the communication delay. The process of fitting a function to the projections of the robot and extrapolating this function, induces large inaccuracies when noise is added to the omni-vision camera data. A new algorithm is proposed which, instead of triangulating the current (extrapolated) omnivision camera data, triangulates past measurements when these are available. After triangulation and filtering by means of a Kalman filter, the ball model is used to predict the 3D ball position at the current time instant. This eliminates the need for extrapolation. The performance of the new algorithm has been assessed using the trajectory generator, results show that this method is more robust for large communication delays in combination with noise on the omni-vision camera data. The new algorithm is implemented on the robots to verify the research above and to provide a base for the future implementation of the algorithm in the robots. The implementation consists out of four parts: a data buffer, a Kalman filter, minimum-distance algorithms and a model update stage. Static tests and dynamic tests verify the implementation of the triangulation algorithm. Static tests achieve an accuracy of < 20 cm. Moreover, the dynamic tests show that a full kick from a robot can be tracked when the frequency of the robot-robot communication is increased. This shows that the triangulation of omni-vision camera data can be used for 3D ball positioning in the MSL league. 3D Ball Positioning by Combination of Omni-vision Camera Data in MSL Soccer Robots iii