Modeling the cerebral architecture of primates is a very active domain of research in autonomous robotics. Indeed, robots can display intelligent complex behaviors if driven by neuronal-like computations that mimic how actions are selected by learning in a behavioral repertoire, from perceptions, goals and past experience. Precise models of the parietal cortex have given efficient sensorimotor coordination capacities, but the same cannot be said about motivations to act. In this domain, reinforcement learning is the main paradigm; when consequences of actions are known, it is a very powerful algorithm to select actions that will maximize rewards, but rewards are here described as a simple scalar qualifying future states. Our claim is that modeling emotional circuits in the brain will endow our robots with richer internal states, as a basis for the motivation to act. In that perspective, we first model respondent conditioning with unconditioned stimuli perceived through pain and pleasure sensors and conditioned stimuli learned by association. In this model of the limbic system and the temporal cortex, a new perception is analyzed through its anticipated consequences on the body and on the external world. We propose that this rich internal state might correspond to an emotional analysis of the world, with three interesting perspectives: (i) to observe in a model the bodily expression of emotions; (ii) to exploit the emotional coloring of the world as an original algorithm for discrimination of perceptions; (iii) to associate this classical conditioning to operant conditioning as a motivational basis to select actions.