State Space Construction by Attention Control

Abstract

In order to understand cognitive aspects of au-tonomous robots, it is fruitful to develop a mechanism by which the robot autonomously analyzes physical sensor data and construct a state space. This paper proposes a coherent approach to constructing such a robot orient-ed state space by statistically analyzing sen-sor patterns and rewards given as the result of task executions. In the state space construc-tion, the robot creates sensor pattern classifiers called Empirically Obtained Perceivers (EOP-s) which, when combined, represent internal states of the robot. A novel feature of this method is that the EOP directs attention to se-lect necessary information, and the state space is obtained with the attention control mech-anism using EOPs. We have confirmed that the robot can effectively construct state spaces through its vision sensor and execute a naviga-tion task with the obtained state spaces in a complicated simulated world. 1