Evolutionary Robot Vision and Fuzzy Evaluation for Natural Communication of Partner Robots

This paper proposes a method of evolutionary robot vision based on a steady-state genetic algorithm and fuzzy evaluation. In order to improve the communication capability of human-friendly partner robots, the perception of human face should be performed as correctly as possible. First, we discuss the concept of evolutionary robot vision in dynamic environments. Next, we propose growing neural gas for preprocessing as a bottom-up processing, and steady- state genetic algorithm for template matching in human face recognition as a top-down processing. In order to improve the performance of the human face recognition, we use fuzzy evaluation for evaluating the degree of human face. Finally, we show several experimental results and discuss the effectiveness of the proposed method. conversation can activate the brain of such elderly people and can improve their concentration and memory abilities. Nursing care for the elderly people can be expected to keep their health by having conversations with robots. However, it is very difficult to continue the meaningful and attractive conversations with robots. Therefore, such a robot requires adaptive perceptual systems to communicate with a human flexibly, and adaptive action systems to learn human behaviors. To realize the learning through interaction with people, we must consider a total architecture of the cognitive development. The cognitive development for robots has been discussed in the fields such as cognitive robotics and embodied cognitive science (8-10). In the previous research of cognitive robotics, many researchers have proposed the learning methods for the achievement of joint attention, imitative learning, linguistic acquisition from the viewpoints of babies and infants (6,10). On the other hand, we focus on the refinement of associative memory by using symbolic information used for utterances and patterns based on visual information through interaction with people as cognitive development of robots. We proposed the concept of structured learning and discussed the importance of total architecture of the learning mechanism (30). However, we did not discuss the performance of the human detection so much. In the previous works, we proposed a simple method of people tracking based on the combination of skin color and hair color, and we have a problem of misdetection of people by objects with similar color combination in the background image. In this paper, we propose a method for detecting a human face based on evolutionary computation and fuzzy evaluation in order to improve the performance of people tracking. The both of evolutionary computation and fuzzy theory are useful and practical in the search under the environment including noise. We proposed the concept of evolutionary robot vision based on the analogy between visual perception and evolutionary search (32). We apply the concept of evolutionary robot vision and fuzzy evaluation for people tracking. The paper is organized as follow. In the section 2, we explain the concept of evolutionary robot vision. Next, in the section 3 we propose a growing neural gas for color extraction and a steady-state genetic algorithm with fuzzy inference for face recognition. Section 4 shows several experimental results and discuss the effectiveness of the proposed method.