Adaptive neural network visual servoing of dual-arm robot for cyclic motion

Purpose - The purpose of this paper is to develop a vision-based dual-arm cyclic motion method, focusing on solving the problems of an uncertain grasp position of the object and the dual-arm joint-angle-drift phenomenon. Design/methodology/approach - A novel cascade control structure is proposed which associates an adaptive neural network with kinematics redundancy optimization. A radial basis function (RBF) neural network in conjunction with a conventional proportional-integral (PI) controller is applied to compensate for the uncertainty of the image Jacobian matrix which includes the estimated grasp position. To avoid the joint-angle-drift phenomenon, a dual neural network (DNN) solver in conjunction with a PI controller and dual-arm-coordinated constraints is applied to optimize the closed-chain kinematics redundancy. Findings - The proposed method was implemented on an industrial robotic MOTOMAN with two 7-degrees of freedom robotic arms. Two experiments of carrying a tray repeatedly and turning a steering wheel were carried out, and the results indicate that the closed-trajectories tracking is achieved successfully both in the image plane and the joint spaces with the uncertain grasp position, which validates the accuracy and realizability of the proposed PI-RBF-DNN control strategy. Originality/value - The adaptive neural network visual servoing method is applied to the dual-arm cyclic motion with the uncertain grasp position of the object. The proposed method enhances the environmental adaptability of a dual-arm robot in a practical manipulation task.