An Approach for Optimal Design of Robot Vision Systems

The accuracy of a robot manipulator’s position in an application environment is dependent on the manufacturing accuracy and the control accuracy. Unfortunately, there always exist both manufacturing error and control error. Calibration is an approach to identifying the accurate geometry of the robot. In general, robots must be calibrated to improve their accuracy. A calibrated robot has a higher absolute positioning accuracy. However, calibration involves robot kinematic modeling, pose measurement, parameter identification and accuracy compensation. These calibrations are hard work and time consuming. For an active vision system, a robot device for controlling the motion of cameras based on visual information, the kinematic calibrations are even more difficult. As a result, even though calibration is fundamental, most existing active vision systems are not accurately calibrated (Shih et al., 1998). To address this problem, many researchers select self-calibration techniques. In this article, we apply a more active approach, that is, we reduce the kinematic errors at the design stage instead of at the calibration stage. Furthermore, we combine the model described in this article with a costtolerance model to implement an optimal design for active vision systems so that they can be used more widely in enterprise. We begin to build the model using the relation between two connecting joint coordinates defined by a DH homogeneous transformation.We then use the differential relationship between these two connecting joint coordinates to extend the model so that it relates the kinematic parameter errors of each link to the pose error of the last link. Given this model, we can implement an algorithm for estimating depth using stereo cameras, extending the model to handle an active stereo vision system. Based on these two models, we have developed a set of C++ class libraries. Using this set of libraries, we can estimate robot pose errors or depth estimation errors based on kinematic errors. Furthermore, we can apply these libraries to find the key factors that affect accuracy. As a result, more reasonable minimum tolerances or manufacturing requirements can be defined so that the manufacturing cost is reduced while retaining relatively high accuracy. Besides providing an approach to find the key factors and best settings of key parameters, we demonstrate how to use a cost-tolerance model to evaluate the settings. In this way, we can implement optimal design for manufacturing(DFM) in enterprises. Because our models are derived from the Denavit-Hartenberg transformation matrix, differential changes for the transformation matrix and link parameters, and the fundamental algorithm for estimating depth using stereo cameras, they are suitable for any manipulator or stereo active vision system. The remainder of this article is organized as follows. Section 2 derives the model for analyzing the effect of parameter errors on robot 3