Robot Calibration: Modeling Measurement and Applications

Abstract

The calibration system proposed showed to improve the robot accuracy to well below 1mm. The system allows a large variation in robot configurations, which is essential to proper calibration. A technique was used and a straightforward convention to build kinematic models for a manipulator was developed, ensuring that no singularities are present in the error model. Mathematical tools were implemented to optimize the kinematic model parameterization, avoiding redundancies between parameters and improving the parameter identification process. A portable, ease of use, speedy and reliable Vision-based measuring system using a single camera and a plane calibration board was developed and tested independently of the robot calibration process. The robot calibration system approach proposed here stood out to be a feasible alternative to the expensive and complex systems available today in the market, using a single camera and showing good accuracy and ease of use and setup. Results showed that the RAC model used (with slight modifications) is not very robust, since even for images filling the entire screen and captured at approximately the same distances from the target, the focus length was not constant and showed an average value shifted by approximately 3% from the exact one. This amount of error can produce 3-D measurement errors much larger than acceptable. Practically speaking, the solution for this problem developed here for a set of camera and lens was to use an external measurement system to calibrate the camera, at least once. The measurement accuracy obtained is comparable to the best found in academic literature for this type of system, with median values of accuracy of approximately 1:3,000 when compared to the distance from the target. However, this accuracy was obtained at considerable larger distances and different camera orientations than usual applications for cameras require, making the system suitable for robotic metrology. For future research it is suggested that the target plate and the calibration board have to be improved to permit the camera to be placed at larger ranges of distances from the target, allowing larger calibration volumes to be used. One path that might be followed is to construct a much larger calibration board, with localized clusters of calibration points of different sizes, instead of just one pattern of point distribution. So, if the camera is placed at a greater distance, larger dots can be used all over the area of the calibration board. If the camera is nearer to the target, smaller dots can be used at particular locations on the calibration board. Different dot sizes make easier for the vision processing software to recognize desired clusters of calibration points. Other sources of lens distortions such as decentering and thin prism can be also modeled, and so their influence on the final measurement accuracy can be understood. Another issue concerns the influence orientation measured data may have on the final accuracy. Non-geometric parameters such as link elasticity, gear elasticity and gear backlash might be modeled, and a larger number of parameters introduced in the model parameterization. This procedure may improve the accuracy substantially if the robot is used with greater payloads.