CPSC 679 Project - Elicitation Process
Development of an Expert Knowledge Base for the Control of an Intelligent Robot Sumo Wrestler
by Vladimir R Vila
Situation Assessment Module
Course Project for CPSC 679 -- Cognitive Processes in Artificial Intelligence
In the case of the RSW application, the issues of positioning, force redirection, attack and defense need to be understood and implemented in the robot's control. These issues have been thoroughly studied and the solutions to the challenges, found in these areas, have been perfected and physically implemented by practitioners of the martial arts. A group of experts from the Chinese combative art, Wing Chuin will be used to compile a rules base that will be used in the robot sumo wrestler's control software. The tactical and eventually the higher level strategic knowledge possessed by these experts will be elicited and used in the fuzzy logic controller.
Challenges Faced in Knowledge Elicitation from Experts
The challenges of knowledge elicitation lies in the diversity of the experts' conceptual constructs and the necessity to use multiple experts because of the distribution of knowledge across a range of experts. It is therefore imperative to have access to a method that addresses the differences in terminologies and conceptual systems so that a unified knowledge base can be rapidly and accurately compiled. Such a system has been developed and documented (Gaines & Shaw, 1989). This system uses an entity-attribute methodology based on Personal Construct Psychology which is designed to allow a significant part of the conceptual systems of experts to be elicited through computer-based interactive interviewing techniques. The system is called the Knowledge Support System Zero (KSS0, Gaines, 1987a; Gaines & Shaw, 1987).
Process of Elicitation
After arriving at a methodology and choosing the tools for the process the following process was implemented:
- Phase 1: A domain or context for the elicitations must first be decided upon. The use of multirelational concept map aid in the visualization and focusing of the domain.
- Phase 2: An initial repertory grid is elicited from each individual expert. Each expert is elicited for their own elements and constructs relevant to the domain. This phase does the double duty of familiarizing the experts with KSS0 and developing an initial set of elements. This initial set of elements will be used to develop the common set of elements that will be used in the other phases.
- Phase 3: All the elements elicited by the experts are combined and discussed. This set is then reduced or increased and changed in any other way come up with a set of elements to fully span the domain being discussed. A second grid is elicited from the experts using the common set of elements while eliciting constructs from each expert. The comparison of the following grids will give allow for the test of Correspondence (the use of different terminology to express the same idea) and Contrast (the use of different terminology to express different ideas).
- Phase 4: Each expert must now rate the constructs of each of the other experts.