Control of a two-wheeled self-balancing robot with support vector regression method

Abstract

Recently, learning based control is a popular topic on robotic applications. This paper presents a novel learning based intelligent control method which realizes the balance control of a statically unstable and dynamically stable robot - a two-wheeled self-balancing robot. The control strategy could be segmented into two levels: a learning based controller using Support Vector Regression approach as a high level and a traditional PD controller as a low level. Support Vector Regression is utilized to learn the mapping between robot's state data and corresponding actions from experiments by using the inclined angle and its angular speed as inputs and the wheels velocity of the robot needed to keep balance as outputs. And the low level PD controller makes sure the motors achieve the velocity value gained before. Experiments are taken to show that the control method is useful and efficient. Additionally, this paper presents a practice of learning based control.