The learning method we develop here applies to an agent trying to learn a function with a continuous domain. We situate the method in the game of robotic soccer.