Incremental knowledge acquisition for complex multi-agent environments

This thesis presents an incremental knowledge acquisition framework that supports the creation of multi-agent teams in complex real-time environments. The underlying technique is based upon the Ripple Down Rules (RDR) technology. RDR is a knowledge acquisition technology that allows the expert to interact directly with the system and construct knowledge incrementally on a case by case basis. RDR and its variants have been successfully applied to a wide range of problem types. However, there has been no research that has explored the adaptation of RDR to complex real-time, multi-agent environments. The domain chosen to develop our system was the Robocup 2D soccer simulation domain. We investigated an SCRDR approach to the problem and from this developed a more complex system based on generalised RDR. This system supported the incremental modelling of intermediate features during the knowledge acquisition process, allowing the experts to create their own abstractions of the domain. The system was extensively evaluated over a period of 6 months, to evaluate the level of performance of the multi-agent teams created by the system and to gather feedback about the usability of the system. During this period of time the system was successfully used by four soccer coaches with differing levels of soccer and computer expertise. All coaches were able to use the system to develop teams that could play at a world class level against the finalists from the Robocup 2007 2D simulation tournament. The approach we present is general enough to be applied to any complex planning problem, with the requirement that a rich feature language is developed to support the specific domain. Our studies confirm that although GRDR seems to be a useful framework for allowing experts to create their own layers of abstractions, in complex domains, some abstractions need to be expressed in low level code below the knowledge level. Our evaluation also suggests that the GRDR system could be improved via further integration with tools to restructure the KB for complex domains where expertise may be experimental and a more radical reorganisation of the KB is required.

Grantor dc:publisher

UNSW, Sydney

Year dc:date

2008

Author dc:creator

• Finlayson, Angela Margaret

• Ripple Down Rules (RDR) technology

Statement dc:rights

• open access
• CC BY-NC-ND 3.0
• free_to_read

Licence

• https://purl.org/coar/access_right/c_abf2
• https://creativecommons.org/licenses/by-nc-nd/3.0/au/

Language dc:language

EN

Identifier

https://doi.org/10.26190/unsworks/19630

OAI identifier oai:identifier

oai:unsworks.library.unsw.edu.au:1959.4/42887