345 
Calibration of CMO-Stereo-Microscopes in a Micro Robot System 
Gaudenz Danuser and Olaf Kübler 
Image Science Laboratory 
Swiss Federal Institute of Technology, CH-8092 Zürich 
phone: +41 1 632 {5281,5282} 
fax: +41 1 632 11 99 
email: {danuser,kubler}@vision.ee.ethz.ch 
KEYWORDS: Light Microscopy, Stereo Vision, Calibration, Bundle Adjustment, Image Distortion, Micro 
Robot Vision 
ABSTRACT 
Micro-Robotics has become an active field of research in the last years. One major problem of this new 
technology is how to build a sensor system that can control the robot motion within the required accuracy. 
Our approach is based on microscopic vision as it has the advantage that the robot and the target objects can 
be observed and positioned within a larger field of view. The strong requirements for accuracy and reliability 
demand a thorough sensor calibration. À versatile photogrammetric approach for three dimensional calibration 
of a Stereo Light Microscope with a Common Main Objective (CMO) lens is proposed. We particularly address 
the problem of how to set up calibration standards with submicron precision and how to compensate the image 
distortion caused by the CMO lens. Numerical results based on simulated data are discussed. 
1 INTRODUCTION 
A new challenge of science and technology in the low micrometer scale is the handling of objects as individ- 
uals, not as ensemble members. Anticipating an emerging importance of micro-technology, the Swiss Federal 
Institute of Technology started a project that aims at the semi automated manipulation of micro-parts with 
sub-micrometer precision. The objectives of a first benchmark are to selectively grasp diamond mono-crystals 
with the dimension of some tens of micrometers, to transport them through a working space of about lcm? and 
to place them in correct orientation with a three-dimensional (3D) precision of about 1um into dimples etched 
on a silicon wafer. To fulfill this task we are developing a six degree of freedom (6DoF) robot system. The 
first experiences with this system will lead us into a wide field of potential applications in micro-machining, 
micro-mechanics and micro-surgery. 
The miniaturization of assembly by robots is accompanied by increasing demands on precision and device 
control. Classical “macroscopic” robots are usually working in an “open-loop” mode. This means, that after 
defining a certain task or sequence of tasks (this can be achieved e.g. by a fixed working program, a scene 
analyzing system or a human operator), the robot is able to carry out the expected functions using its own local 
sensors. Based on a thorough calibration, the various sensor informations can be transformed into an absolute 
world coordinate frame. This approach is based on temporal and random errors that are below the specified 
working precision. Both assumptions are not valid in the case of micro-robotics. Therefore, it is imperative to 
use global sensors which are able to measure the geometrical relation between the target object and the robot 
tools within a larger field of view (FoV). An ideal global sensor would provide 6D (position and orientation) 
data of the entire working space on the required precision level and in real-time. Unfortunately, this type of 
sensor does not exist. 
To minimize the trade-off between the size of the FoV, the speed and the precision we propose to employ stereo 
light microscopic vision. Stereo microscopes deliver 3D data of the micro-world and, compared to any other 
sensor system, have by far the largest FoV per time unit. The price for this is the high computation cost, as the 
data for the robot control must be extracted from stereo image sequences. À second problem is that occlusions 
may cause data drop-outs during the grasping procedure of the robot. For that reason we decided to add some 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop “From Pixels to Sequences”, Zurich, March 22-24 1995