Integrating perception, action and learning

Abstract

The construction of robots adequate for field operation requires the integration of planning, perception, and learning. The coordination of these technologies in the context of constraints of realtime, robust, and safe operation is leading to coupled representations of perception and action. We are developing a system that explores the use of Case-Based Reasoning techniques for inducing low-level, coupled representations of potential robot locations. These will subsequently be used for planning and possibly as operationality criteria in deriving efficient task description. This work is proceeding under an architectural design that integrates horizontal, subsumption approaches to building robots with the vertical Perception-Control-Action internal structure for each layer.