A reinforcement learning approach to score goals in RoboCup 3D soccer simulation for nao humanoid robot

Reinforcement learning is one of the best methods to train autonomous robots. Using this method, a robot can learn to make optimal decisions without detailed programming and hard coded instructions. So, this method is useful for learning complex robotic behaviors. For example, in RoboCup competitions this method will be very useful in learning different behaviors. We propose a method for training a robot to score a goal from anywhere on the field by one or more kicks. Using reinforcement learning, Nao robot will learn the optimal policy to kick towards desired points correctly. Learning process is done in two phases. In the first phase, Nao learns to kick such that the ball goes more distance with minimum divergence from the desired path. In the second phase, the robot learns an optimal policy to score a goal by one or more kicks. Using this method, our robot performance increased significantly compared with kicking towards predetermined points in the goal.