In this Chapter, we imagine the Field Robot as an observer of the signals received from its sensors. As the signals are corrupted by noise, uncertainty about the meaning and the origin of the signals needs to be captured within the algorithms that generate actuations given the input sensor signals. There are different ways to describe and process this uncertainty, and in the first part of this chapter we list the frequentist’s, the Bayesian, and Dempster-Shafer’s approaches. In the second part of this chapter, we look at ways how to combine the three different formulations. We introduce the FISST, BGoF and Blind Separation methods. In the third part, we apply the three different ways to Decision-Making tasks, finding that the three approaches fit to different settings: frequentists for fusion with no prior information, Bayesian for fusion with prior information, Dempster-Shafer for fusion with inconsistencies. The Field Robot will need a mixture of these views: as less data as possible, as less prior information as possible, as less inconsistencies as possible. Hence, the best Field Robot Design has to apply a sequence/mixture of FISST, BGoF, and Blind Separation on the incoming data to minimize the amount of data (resulting in quick reaction), the amount of pre-programming (resulting in higher autonomy), and the value associated to inconsistency (resulting in higher robustness).