The project proposes the definition of a new kind of data-driven control of robotic platforms, especially relevant for environmental observation. This new approach to the control of robotic platforms extends the recent paradigm of data driven control (that considers control problems whose objective function is directly defined in the sensor space), by considering objective functions that are defined in the space of global predictions of the field of interest. This means that rather than trying to maintain or attain a given representation from the sensor's outputs, these techniques aim at achieving in the most efficient way, a given confidence in a global internal representation of the observed field.