Emotions are complex phenomena that involve a number of related subsystems and can be activated by any one (or by several) of them. Elicitors of emotions are for example grouped in  under the categories of neuro-chemical, sensorimotor, motivational, and cognitive. Some of these elicitors are emotion-specific, but emotions also show a certain degree of generality  (e.g., of object, of time) that accounts for the fact that a person can experience the same emotion under different circumstances and with different objects. But if emotions show this generality, what accounts for the activation of different affects? As we already mentioned, activation theories that only take into account the arousal and valence properties of emotions, are not able to fully account for their differential activation. To overcome this problem, Tomkins  proposed three variants of a single principle:
The fact that this model is based on different patterns of stimulation makes it particularly suited for emotion activation in our robot, since we were aiming at physical interaction based on tactile stimulation. However, an analysis of the possible interactions that humans could have with Feelix revealed two cases that the model did not account for. On the one hand, the model did not propose any principle to activate negative emotions such as sadness or boredom. On the other hand, happiness is only considered in the sense of relief resulting from stopping a very high (and therefore annoying) level of stimulation, i.e., as the cessation of a negative stimulus. Happiness, however, can also be produced by positive stimuli, such as gentle interaction. We have therefore refined the model by postulating two more principles:
One of the drawbacks of this model lies in the fact that, being of general nature rather than associated with specific stimuli, some of the patterns can activate more than one emotion; another element is therefore needed to differentiate among emotions activated by the same general pattern. In the simulated creatures presented in , this was achieved by means of a synthetic physiology that allowed to associate to basic emotions a set of physiological parameters specific to each of them (e.g., fear is characterized by high heart rate and low skin temperature, versus interest that is accompanied by low heart rate).Since we wanted to do all the computations on board, simulating a rich enough physiology was not appropriate. We therefore used a simpler discrimination criterion based on the intensity with which each pattern occurs (e.g., the stimulation increase is higher in the case of fear than in that of surprise). This criterion was sufficient to characterize the subset of emotions we used in Feelix.