EMOTIONAL EMPATHY MODEL FOR ROBOT PARTNERS USING RECURRENT SPIKING NEURAL NETWORK MODEL WITH HEBBIAN-LMS LEARNING

This paper discusses the development of an emotion model for robot partner system. In our previous studies, we have focused only on the robot's emotional state. However, the emotional state of the other party is also an important factor for smooth conversation in human society. Therefore, the robot partner has two emotional structures for human: empathy and robot emotion. First, human empathy uses a perceptual based emotion model to know the human's emotional state based on the sensory information. Next, we propose a recurrent simple spike response model to improve the robot's emotional model, and we apply "Hebbian-LMS" learning to modify the weights in the spiking neural network. The robot's emotional state is calculated by using the human's emotional information, internal and external information. The robot partner can use the emotional results to control the facial and gesture expression. The utterance style is also changed by the robot's emotional state. As a result, the robot partner can interact emotionally and naturally with human. First, we explain the related works and the development of the robot partner "iPhonoid-C". Next, we define the architecture of the emotional model to realize emotional empathy towards human. Then, we discuss the algorithms and the methods for developing the emotional model. Finally, we show experimental results of the proposed method, and discuss the effectiveness of the proposed structure.