ADAPTABILITY TO PERIODIC VARIABLE DISTURBANCE USING PROBABILISTIC STATE-ACTION PAIR PREDICTION

When operating a robot in a real environment, its behavior is probabilistic because of slight transition of the robot’s state or error in the action taken at a given time. In this case, it is difficult to operate the robot using rule-based-like action decision methods. Therefore, ad-hoc-like action decision methods are needed. A method is proposed for deciding on future actions based on a robot’s present information. The state-action pair prediction method has been reported; it links the state and future actions of a robot using internal information. A statistical approach to state-action pair prediction has been introduced previously, in which the existence probability of a state and action in the future is calculated according to the normal distribution. This paper considers the situation where a command input is sent to an inverted pendulum. Based on this command input, the shape of the floor is changed from flat to undulating. The results of verification experiments confirm that the proposed method can adjust the shape of the floor autonomously.