An autonomous robot that eats information via interaction with humans and environments

Abstract

We describe a behavior control architecture, EGO, for autonomous robots, by which robot behaviours are selected to keep internal variables in proper ranges. The EGO is based on ethological study, so that both external stimuli and internal variables cause its behavior selection. In addition, introducing "remaining information" to one of the internal variables, we can develop "information eating" behavior of the autonomous robot via interaction with humans and environment, which is actually symbol acquisition behavior. We propose the concept of an emotionally grounded symbol, which gives information about the importance of the object for the survivability of an autonomous robot. In addition an emotionally grounded symbol gives information about emotions that were experienced during learning as well.