Pose Optimization of a Single Ball Self-Balancing Mobile Robot Based on IMU and Visual Fusion

Abstract

The single ball self-balancing mobile robot is a typical multivariable, high-order, nonlinear and static instability system, IMU inertial devices are generally used to obtain attitude parameters of the system. The data errors obtained by the inertial devices will accumulate over time, and the long-term pose estimation is rather unreliable. But for some quick motion over a short period of time, inertial sensors can get very accurate estimates. Monocular cameras are often used as image real-time monitoring sensors by single ball self-balancing mobile robots. The camera's data does not accumulate errors over time, so there is essentially no drift. Therefore, camera pose estimation can effectively correct the inertial sensor error. In a fast-moving environment, the shutter-roller camera cannot capture the image clearly and there may be ghosting or overlapping of two areas too little, resulting in mismatched or unmatched conditions. So that the robot makes a wrong estimate. In this paper, the non-linear optimization method is used to fuse the IMU and monocular attitude data to obtain the optimal pose estimation of the robot. Based on the modeling of robot using Lagrange's equation method, the control of single ball robot is realized by using fuzzy PID algorithm. Experiments show that this scheme can respond quickly to the change of the attitude of the ball self-balancing robot and keep the attitude of the robot relatively stationary, so that the robot can move steadily.