Robot self-awareness: Exploration of internal states

Abstract

A self-aware system has the possibility of dealing with novel situations more effectively than a system without self-awareness. A selfaware system can attend to its own internal states, thus providing a means of generating introspection and self-modification capabilities. A robot needs a capability to attend to its internal states in order to be genuine self-aware. Internal states can be made up of emotion, belief, desire, intention and expectation or it can be processes such as sensation, perception, conception, simulation, action, planning and thought. It is crucially important to be aware of its own emotions, perceptions, beliefs and intentions during the recognition process. Currently, developments in the field of self-awareness of robots are mainly based on a mimicry of human internal states. It is difficult for systems developers to specify specific internal states for all possible conditions and situations. It is obvious that such systems have very limited opportunities for self-development. In this paper we consider an approach that allows the robot to generate their own internal states. These internal states are not similar to the human internal states. Such property gives the system of internal states plenty room for self-development. We propose a new model of genetic algorithm for analysis of robot control system and generation of new internal states.