Phylogenetic and ontogenetic learning in a colony of interacting robots

The objective of this paper is to describe the development of a specific theory of interactions and learning among multiple robots performing certain tasks. One of the primary objectives of the research was to study the feasibility of a robot colony in achieving global objectives, when each individual robot is provided only with local goals and local information. In order to achieve this objective the paper introduces a novel cognitive architecture for the individual behavior of robots in a colony. Experimental investigation of the properties of the colony demonstrates its ability to achieve global goals, such as the gathering of objects, and to improve its performance as a result of learning, without explicit instructions for cooperation. Since this architecture is based on representation of the ''likes'' and ''dislikes'' of the robots, it is called the Tropism System Cognitive Architecture. This paper addresses learning in the framework of the cognitive architecture, specifically, phylogenetic and ontogenetic learning by the robots. The results show that learning is indeed possible with the Tropism Architecture, that the ability of a simulated robot colony to perform a gathering task improves with practice and that it can further improve with evolution over successive generations. Experimental results also show that the variability of the results decreases over successive generations.