This paper presents the theoretical framework for the centralized control architecture of the multi agent microrobotic platform MiCRoN. The entire control system architecture integrates sensory modules, modeling modules, and control modules. The latter are composed by (i) a high level simulation and autonomous execution unit that is capable for on-line multi-robot navigation with collision avoidance, (ii) a trajectory tracking unit for manipulation purposes, and (iii) a low level position controller that performs position control exploiting machine learning algorithms. The high level controllers take into account behaviors specific to the micro-scale. The performance of the layered control system is evaluated through simulations and preliminary hardware experiments on a micro-robotic platform. The application domain of the MiCRoN platform is cell manipulation, and 3-D assembly for micro-fabrication.