Behavioral Development of Ball Kicking Motion of a Two-wheeled Inverted Pendulum Mobile Robot

In this paper, we introduce a method for generating a dynamic motion such that a two-wheeled inverted pendulum robot kicks a ball far away utilizing its own body dynamics while it keeps standing. Such a dynamic motion can be acquired through trial and error based on a reinforcement learning scheme. We utilize a simple policy gradient method to acquire a kicking motion which is designed by defining the desired parameters such as body angle, wheel angular velocity and so on. To show the validity of our approach, we perform computer simulation experiments of behavior acquisition for the two-wheeled inverted pendulum robot. Based our approach, we succeeded in acquiring the kicking motion of the two-wheeled inverted pendulum robot. A very interesting finding is: each of the acquired motions deviates from the desired trajectory, which is given by the human designer while keeping evaluation value of the acquired motion high.