Design of a visual control based using optical flow for a ball-riding robot

Abstract

Characteristic of the ball-riding robot is always fascinating humans. It can be developed for serving human in several applications. In order to operate the ball-riding robot for the intelligent tasks, an image processing and computer vision are required to be implemented into the robot system while stabilization is still needed. To prevent the cause of serious issue for implementing into a robot, the simulation is best testing tool that can evaluate robot performance before conduct a real implementation. A visual odometry estimation via homography is used to estimate the robot leaning angles, and used as a feedback data for a robot balancing controller. In this research, the simulation is implemented in Gazebo simulator, the ball-riding robot model is added into the Gazebo environment with a RGB camera, Inertial Measurement Unit (IMU) and quadrature encoder sensor. The low-level LQR+I controller and the visual odometry estimation are implemented into simulator by using Robot Operating System (ROS) node and communicating with Gazebo via publisher/subscriber protocol.