Animal-like control of intelligent autonomous agents

This paper reviews key concepts of CZAR (Czech Animal-Like Robot) architecture. This hybrid autonomous agent architecture was designed for the usage mainly in the A-Life domain and combines knowledge-based and behavior-based approaches. Its structure, strengths as well as weaknesses, and roots in biology are presented. CZAR has arisen as a result of a number of applications, where real robots with variety of control systems were tested and several biological systems were simulated. During this iteration process, basic principles common to mobile robotic applications and common to real animals were identified and embodied in the resulting architecture. This article highlights these principles and provides detailed description of the CZAR. Basic building blocks include perception, actuation, internal state model, cognition, action selection and a special importance is put onto attention selection. Particular focus is also placed on learning and adaptivity. A simplified ecosystem was created to test CZAR.