Adaptive Visual Servoing Control of robot Manipulator for Trajectory Tracking tasks in 3D Space

Abstract

This paper presents an adaptive visual servoing controller for trajectory tracking in the robotic manipulators. The proposed controller is based on the inverse kinematic model of a robotic manipulator (Arm-6A×18) and position-based visual servoing control. In this research work a Kinect camera, which is fixed as an eye-to-hand configuration, is used for extracting the trajectory in the color and depth images. Using this information, the geometric coordinates of points in Kinect coordinate space are calculated. Afterwards, these coordinates are converted to the robot's coordinate space using an obtained transformation matrix. The parameters of this matrix are calculated using Linear Least Squares Estimator (LSE) method. Then, the inverse kinematic model of manipulator is applied on these coordinates to determine the angle of each robot's joint. Due to the brightness intensity and other uncertainties, the Kinect camera has error in the calculation of depths. To cope this problem, an adaptive learning algorithm, called Weighted Recursive Least Square Estimator (WRLSE), is applied. This algorithm adaptively tunes the parameters of transformation matrix in order to reduce the distances between end-effector's trajectory and reference trajectory. For tracking the reference trajectory, Canny edge detector is used to detect the trajectory in the image. Thereafter, the detected trajectory is discretized to the set of target points which should be reached by the end-effector. The proposed adaptive controller is applied on a real robotic manipulator. The experimental results show that by using the adaptive learning algorithm, the end-effector can track the trajectory with high accuracy.