Building complete autonomous agents: a case study on categorization

Abstract

With the advent of new artificial intelligence it has become clear that in order to understand intelligence it is important to build complete agents. A complete agent is capable of behaving autonomously in an environment without human intermediary. It has to incorporate, among other things, capabilities for categorization, for navigation, and for "deciding" what to do. We illustrate how such an integration can be implemented on a mobile robot. We introduce a new control architecture, the so-called Extended Braitenberg Architecture (EBA). It consists of loosely coupled processes that run in parallel. They implement various behaviors such as exploration or recharging. The task of the robot is to collect certain types of objects in the environment. We show how category learning can be embedded in the overall architecture, i.e. we illustrate how categorization can be viewed from a complete agent's perspective. Results on the behavioral performance as well as the underlying internal dynamics are presented.