Nanotechnology is one of the most active research fields all over the world. Anticipating the emerging importance and the great challenges that are coupled with the miniaturization of sensing, handling and assembling structures on the nanometer scale, an ETH project was started with the goal to construct a nanorobot system. For many reasons, microscope imagery is the most appropriate data from which one can extract feedback information for robot control. In our project, we employ a stereo light microscope. New photogrammetric and computer vision algorithms have to be derived to provide the necessary geometric control data to the robot. Contributions to three logically connected problems form this thesis: The geometric calibration of the light microscope is the key to derive 3D information with the required precision. Subsequently, the target objects and the robot manipulation tools have to be reconstructed from stereoscopic cues. Finally, vision controlled grasps should be performed. This proposal addresses the key issues of these three steps and sketches the strategies that are intended do be implemented. The first two sections contain a detailed description of our approach. This provides an overview of the whole project, not only the computer vision aspects, and explains the significance of microscopic stereo vision in particular. In the last section I elaborate on the significance of the proposed thesis for academic and industrial applications.