Experientially Guided Robots

Abstract

This paper argues that an experientially guided robot Is necessary to successfully explore far-away planets. Such a robot is characterized as having sense organs which receive sensory information from its environment and motor systems which allow it to interact with that environment. The sensori-motor information which it receives is organized into an experiential knowledge structure and this knowledge in turn is used to guide the robot's future actions. A summary is presented of a problem solving system which is being used as a test bed for developing such a robot. The robot currently engages in the behaviors of visual tracking, focusing down, and looking around in a simulated Martian landscape. Finally, some unsolved problems are outlined whose solutions are necessary before an experientially guided robot can be produced. These problems center around organizing the motivational and memory structure of the robot and understanding its high-level control mechanisms. This paper discusses a project which is attempting to develop the "mind" of' a robot which will be capable of experiencing its environment, storing sensori-motor information, and then using its accumulated knowledge to guide its future actions. In Section I, we describe the sort of behavior that an experientially guided robot might exhibit, and we give some reasons why we might want such a thing. Then, in Section II, we discuss the current state of our project, and in Section III we indicate some important issues that are as yet unresolved.