Optical tweezers offer certain advantages such as multiplexing using a programmable spatial light modulator, flexibility in the choice of the manipulated object and the manipulation medium, precise control, easy object release, and minimal object damage. However, automated manipulation of multiple objects in parallel, which is essential for efficient and reliable formation of micro-scale assembly structures, poses a difficult challenge. There are two primary research issues in addressing this challenge. First, the presence of stochastic Langevin force giving rise to Brownian motion requires motion control for all the manipulated objects at fast rates of several Hz. Second, the object dynamics is non-linear and even difficult to represent analytically due to the interaction of multiple optical traps that are manipulating neighboring objects. As a result, automated controllers have not been realized for tens of objects, particularly with three dimensional motions with guaranteed collision avoidances. In this paper, we model the effect of interacting optical traps on microspheres with significant Brownian motions in stationary fluid media, and develop simplified state-space representations. These representations are used to design a model predictive controller to coordinate the motions of several spheres in real time. Preliminary experiments demonstrate the utility of the controller in automatically forming desired arrangements of varying configurations starting with randomly dispersed microspheres.