The state of the art of perception processes for the autonomy of robots is constantly improving, yet these processes remain mostly pre-configured at the robot design phase. This prevents their adaption to the context at hand, which is all the more needed for long life systems that encounter a large variety of situations. This paper presents work on the modelling of perception processes, exhibiting the need to assess their quality, so as to be able to actively control them. We instantiate the visual odom-etry case, a crucial functionality for planetary rovers, and define dedicated data quality assessment functions for the elementary processes composing it. These functions are used to monitor the processes, defining control points onto which explore different parameter configurations that better adapt to the context the robot is facing. Preliminary tests are performed using a planetary analogue data set to show the potential of this approach.