Minoru Asada and Takashi Minato
Perceptual attention can be regraded as the first step towards symbol emergence from senroy data. Especially, visual attention is one of the key issues for robots to accomplish the given tasks, and the existing methods specify the image features and attention control scheme in advance according to the task and the robot. However, in order to cope with environmental changes and/or task variations, the robot should construct its own attention mechanism. This paper presents a method for image feature generation by visio-motor map learning for a mobile robot. The teaching data constructs the visio-motor mapping that constrains the image feature generation and state vector estimation as well. The resultant projection matrix from the filtered image to a state vector tells us which part of the image is more informative for decision making than others. The method is applied to indoor navigation and soccer shooting tasks, and discussion is given.