#ORF09 Designing a System of Rule Based Agents Presentation

This entry is part 19 of 30 in the series October Rules Fest 2009

Luke Voss is back presenting on rule based agents.

“We need more software that does work”

This is what Luke is really working on these days. It was inspired by his work at NASA’s Jet Propulsion Laboratory in the Advanced Concepts Design Team (Team X).

Team X has design sessions (3 x 3 hours sessions over a week) and uses a process where they:

  • Identify Science goals
  • Identify Mission Constraints
  • Explore a baseline design
  • Time-permitting, explore alternatives
  • Draft a final report

The models an tools that they use are well known because space missions have the same basic structure. One of the things they use is a set of network spreadsheets that share information. The higher level spreadsheet gathers the information from sub-spreadsheets to have a more complete view. Not very high tech, but it apparently works for their purpose.

This way of working, gave an large improvement over traditional design sessions. Is there a way to get more improvements (time, costs, quality, quantity)? Some of the ideas being discussed to get this are:

  • Compositional Models
  • Altered team structures (simplified teams with simpler systems kind of kept static)
  • Executable models
  • Rule-based system

It was obvious to Luke that the rule based  approach had a lot of potential. He is thinking about:

  • Agent-based
    • Networked
    • Modeled after the existing team
  • Two modes
    • Solution path (to get a baseline design)
    • Tradespace exploration (to get a list of designs)

The Agents can be broken down into different types covering a spectrum from “Fuzzy” to “Well-defined”:

  • Coordination Agent
  • Domain Expert Agent
  • Algorithmic Agent
  • Human Agent

Why an Agent based approach as opposed to a single rulebase? Each agent will have relatively well defined boundaries which makes it easier to test an agent and to understand it. But it makes troubleshooting of the interactions more complicated. The agents enforce a separation of concerns. Crossing the boundaries of an agent is more costly, but the trade-off is worth it. Maintenance of the rules is easier because of the limited scope of the rules. The performance is not as important a concern since computer time is cheap when compared to human time.

This is about automating boring engineering.

Very interesting talk with a practical outcome and a lot of potential to help the engineering field.

Series Navigation<< #ORF09 Measuring your Rules’ KPI Presentation#ORF09 Extending General Purpose Engines Presentation >>

Leave a Reply

Your email address will not be published. Required fields are marked *

*


*

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>