Computational emotion models for virtual humans are very important in many fields. A believable virtual human should have both geometry shape

and autonomous emotion. Based on previous researches on emotion models

a new motivation-driven based autonomous emotion model is set up. A cognitive architecture of virtual human is presented by psychology and neural science; it includes perception

emotion

mood

personality

motivation and behavior. A formalization of inner motivation and external stimuli is built; the intensity of a basic emotion can be calculated in the model

and emotion state is the union of all basic emotions. In order to realize the self-adapting mechanism of emotion

a Q-learning method for emotion learning is also proposed. A 3D virtual scene with virtual humans is realized on PC

new emotion model can make a virtual human’s autonomous emotion expression more believable

the experiment shows that our method is effective in virtual human’s emotion simulation.