IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 62, NO. 5, MAY 2015 2921 Calibration-Based Iterative Learning Control for Path Tracking of Industrial Robots Yi Min Zhao, Yu Lin, Fengfeng Xi, and Shuai Guo Abstract —This paper addresses the problem of path tracking of industrial robots. The main idea is to correct a preplanned path through an iterative learning control (ILC) method. Instead of seeking the conventional ILC strategy, an iterative learning identification method, which is called calibration-based ILC, is developed to identify the robot kinematic parameters along the path in a local working zone. To facilitate calibration-based ILC, we propose two objectives. The first objective is to find the exact values of robot kinematic parameters based on the ILC scheme. The second objective is to search the fastest learning conver- gence speed and robustness in the iterative domain. Based on the identification of robot kinematic parameters, we then propose an algorithm for the accurate path tracking of industrial robots. The simulation and experimental results demonstrate that the performance of path tracking can be improved significantly via the proposed method. Index Terms—Iterative learning control (ILC), path cor- rection, path tracking, robot calibration, visual servoing. I. I NTRODUCTION P ATH TRACKING of industrial robots aims at accurate path positioning along predefined paths in the robot workspace. Path tracking involves in a number of industrial robot applications, such as riveting, welding, painting process, material handling, part assembly, etc. Recently, high accurate path tracking is increasingly demanded in most applications of industrial robots. For instance, in aerospace assembly, riveting has been considered as one of the major joining methods [1]. Improved performance of the riveting process is becoming necessary because of competitive markets. Over the last number of years, robots have been used to provide panel holding and feeding functions for conventional squeezing riveting machines [2]. Recently, Xi et al. [3] have developed a novel robotic riveting system with a robot holding a percussive riveting gun equipped with a rivet feeder. There are two separate steps in this robotic riveting process: hole drilling and rivet-in-hole insertion. In practice, there always exist undesired errors, which Manuscript received March 10, 2014; revised July 25, 2014 and September 17, 2014; accepted September 30, 2014. Date of publication October 24, 2014; date of current version April 8, 2015. Y. M. Zhao is with the College of Engineering and Information Tech- nology, University of Arkansas at Little Rock, Little Rock, AR 72204 USA (e-mail: ymzhao@ualr.edu). Y. Lin and F. Xi are with the Department of Aerospace Engi- neering, Ryerson University, Toronto, ON M5B 2K3, Canada (e-mail: yu.lin@ryerson.ca; fengxi@ryerson.ca). S. Guo is with the School of Mechanical Engineering, Shanghai University, Shanghai 200336, China (e-mail: guoshuai@shu.edu.cn). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TIE.2014.2364800 can be classified into two categories: 1) intrinsic errors such as geometric parameter variation in the robot/tooling kinematic model due to manufacturing tolerances, assembly clearances, mechanical flexibility, etc., and 2) extrinsic errors such as exter- nal disturbances. In reality, the repeatability of industrial robots can be as high as 0.02 mm, whereas their position/tracking accuracy can be as low as 1 mm [4]. The underlying problem is that, although an industrial robot is calibrated, it is calibrated over its entire workspace. As a result, the calibrated kinematic parameters are generally the averaged values of those calibrated over the robot workspace. In other words, a calibrated robot cannot guarantee absolute positioning/tracking accuracy at any given points. This is equivalent to saying that kinematic param- eter variation always exists in the calibrated robot kinematic model, which results in positioning/tracking errors. Therefore, the arising path-tracking issue of robotic riveting is that, al- though the holes are drilled according to a planned path, rivets cannot be inserted into individual holes along the same path accurately due to the presence of the aforementioned errors. Thus, a robust controller for accurate path tracking is the main challenge in the design of the robotic riveting system. Various researches on path-tracking control strategies for robot manipulators have been studied [5]–[12]. Most are based on knowing the robot’s governing dynamics and applying the path constraint to those dynamics to define a lower dimensional dynamic system [6]–[8], [11], [12], and the path-tracking input of these control schemes is the torque in the robot joint space. In [5], ultrasound position sensors are used to measure path offsets between the idealized and the actual location of the robot end-effector and then provide the path-correcting control input to the robot controller with robot joint coordinates. In [10], the direct learning control strategy is used for path tracking by the learned feedforward control programs over a domain of robot joint parameters. Since an industrial robot controller is not open, these path-tracking schemes cannot be exploited in path tracking of industrial robots. Therefore, an efficient and robust control scheme should be explored for the path tracking of industrial robots without the need to access the robot’s internal control unit. Nowadays, iterative learning control (ILC) has become one of the most effective control methodologies in dealing with repeated tracking control problems or periodic disturbance rejection problems [13], [14]. The key design feature of ILC is the efficient use of past information to improve tracking performance within a small number of trials, while ensuring robustness of the process against system uncertainty [15]–[17]. Many ILC algorithms have been proposed to design a purely feedforward action depending solely on the previous control 0278-0046 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.