KNOWLEDGE-BASED SYSTEM FRAMEWORK FOR ENVIRONMENTAL PERCEPTION IN A SUBSEA ROBOTICS CONTEXT.

A knowledge-based system framework is described that provides the infrastructure necessary to combine the adaptive guidance and control, sonar perpection, 3-D world interpretation, and man-machine interface functions for an autonomous submersible. The component parts of such a robotic system are discussed, followed by a description of the information and processing subsystems that are required. The framework within which they are implemented uses the architecture of a blackboard system to specify a modular unit (the blackboard cell) capable of manipulating the diverse types of process and uncertain information encountered. Interconnection of these modular units uses a standard interface configured as a globally accessible blackboard. As a means of evaluation, a software implementation of the architecture is applied to the sonar perception subsystem. The image processing tools at the heart of a simple knowledge base are described. The results show a small sequence of system execution, and some examples of candidate object and shadow areas derived from real sonar images are presented. It is demonstrated that the blackboard cell allows a rich feedback structure to the flow of information and processes. The feedback structure allows both model and data-driven actions to support the formation of competitive and cooperative solutions.