The main goal of formation control strategies is to coordinate a robot team to achieve a desired formation pattern. Some applications of formation control, for instance sensing coverage, searching and rescue, transportation of large objects, etc. require to establish the desired positions of the robots according to an strategic and equidistant spatial coverage within the work area or boundary. This paper presents a hybrid architecture where a SOM neural network establishes the strategic positions of a 2D area or perimeter using the formation graph of robots. This information is transmitted online to a low-level control strategy based on artificial potential functions which ensures the convergence to the desired formation and collision avoidance based on decentralized repulsive vector fields instead the common repulsive potential functions. Some numerical simulations with virtual reality show the performance of the control architecture.