Content area
Full Text
1. Introduction
Robotic manipulators are widely used for industrial tasks such as part assembling, welding, etc. However, most of them are used by teaching playback strategy, i.e. the engineer directly operates the manipulator first to accomplish the task and then the robot replays the trajectory of joints to perform a repetitive task. It takes great effort to operate and program the robot, and the system needs to be retaught if the pose of the target changes. To this end, vision-based methods were proposed a few decades ago (Shirai and Inoue, 1973); vision-based control methods use visual information instead of commonly used physical sensor signals in the control loop, achieving good flexibility to perform complicated tasks.
Vision-based control methods can be divided into three kinds, i.e. position-based, image-based and hybrid methods. Recent reviews are Chaumette and Hutchinson, 2006, 2007. By comparison, image-based and hybrid methods are more robust to environment uncertainties and rely less on the accuracy of the vision system than position-based methods (Chaumette, 1998). Generally, most image-based visual servo systems use monocular cameras, without the need of the expensive and carefully calibrated stereo vision system. However, due to the loss of depth information, model uncertainties exist in these systems and affect the stability of the control system. Various adaptive control methods are proposed to deal with model uncertainties, and online adaptive parameter identification is used to guarantee the stability of the control loop (Chen et al. , 2005). Besides, some robust control methods are also used such as sliding mode control (Becerra et al. , 2011), mixed H 2 / H [infinity] control (Tu and Ho, 2011), etc. However, most of the existing methods are designed for set-point control task, and suffer weak dynamic performance when performing a trajectory tracking task due to the loss of depth information. For adaptive control methods, tracking performance is limited by the convergence of parameter identification.
Iterative learning control is widely used in industrial repetitive tasks, and precise trajectory tracking can be achieved by iterative learning. In (Jiang and Unbehauen, 2002; Jiang et al. , 2007), iterative learning control is used for visual trajectory tracking of eye-to-hand systems. In (Jiang and Unbehauen, 2002), a filter-based iterative learning method is proposed for a general nonlinear model, image feature...