Transcript in design
PACT and RAPID: Multi-Agent Design and Concurrent Engineering by Greg Milette CS525M, Spring 2002 The Plan •Describe the problem. •Describe how PACT solves the problem. •Describe how RAPPID solves the problem. (23 slides) Concurrent Engineering The Problem •Each engineering group is specialized but must coordinate designs. •Different engineering groups use their own specialized design tools. •Each with different assumptions, conventions, and limitations. •Different tools are highly specialized and cannot be to converted. Solutions •Allow groups to use their own specialized tools, while enabling the groups to reason collectively about design. PACT is -A system designed to integrate existing multi-tool systems. -Uses agent communication and common ontology. RAPPID is -A system design to reason about trade-offs. -Using market dynamics and set-based reasoning. PACT: The Agents for the Design of a Manipulator Sharing design models A) could be hard to scale B) scales well but requires: •Way to coordinate interactions (Facilitators) •Way to express design decisions (Agent Messages) Advantage of Sharing Design Model •A unified model is not needed. Instead tool models are encapsulated. •Shared engineering language is needed for communication. It only has to cover a the shared design model •Communication is formally defined, through KIF and KQML, which increases modularity. Facilitators •The four specific functions a facilitator provides: •Translates tool-specific knowledge into and out of standard knowledge exchange language. •Provides a layer of reliable message passing. •Routs outgoing messages to appropriate destination. •Initializes and monitors the execution. Shared Model Agent’s Model Facilitator Agent Ontology •Was needed to express engineering concepts. •Example concept: voltage on a wire. •What units? •What time interval discrete or continuous? •Better to agree on a method for describing information. Do not standardize. Example sentences using ontology: (physical-dimension length) (unit-of-measure inch) (quantity (diameter shaft-a)) (= (diameter shaft-a) How to Develop an Ontology •Developers met and role-played how their components would respond •Then corresponded by email. •But this would not work in a larger project, which need standard ontologies. Example of Distributed ReDesign Assert part number #1234. Dimensions 3.0 x 3.0 meters. Digital Circuitry + component catalog Control Adjust design Assert part number #1234. Dimensions ?x. Mechanism Power System and Sensors Assert new motor part # 1234 PACT •Model – each agent models the world in it’s own way, it will communicate its model through a shared language. •Agent structure: •Each agent can be in a different physical location. •Federation architecture through facilitators. •Communication – KQML, KIF, TCP/IP and email messaging •Ontology – allows engineering concepts to be transferred What is RAPPID? •Agents can reason about the trade-offs among design options using set-based reasoning. •Example of trade-off: How much power should be budgeted for sensor circuitry? And the actuator? •Can solve a big optimization problem •Consider many design alternatives automatically. Two types of agents •Components – •Represents a part of the design. •Buys and sells characteristics in the market. •Might be organized in a hierarchy. •Might be controlled by a human user. •Characteristics – •Definable attribute like weight or power. •Maintains a marketplace for that item. The RAPPID System What is Set-Based Reasoning? http://www.erim.org/cec/rappid/rappid.htm Set-Based Reasoning in Context •Low prices = slack characteristics •High prices = constrained characteristics •Space can be collapsed by buying up allocations of characteristics… •Which gives other agents more funds to purchase other characteristics instead… •Which causes the amount of certain characteristics to get fewer and converge on a price. Set-Based Reasoning Reasoning About Bids Conclusions •In concurrent engineering groups: •Are isolated because they use different techniques and models to solve problems. •But they need to reason collectively. •PACT: Use communication and a shared ontology to exchange design decisions. •RAPPID: Use market dynamics and set-based reasoning so agents can reason about the trade-offs of design choices. References •M. R. Cutkosky, R. S. Englemore, R. E. Fikes, T. R. Gruber, M. R. Genesereth, W. S. Mark, J. M. Tenenbaum, and J. C. Weber. PACT: An Experiment in Integrating Concurrent Engineering Systems. IEEE Computer, 26 (Januarty)(1):28-37,1993 •H. V. D. Parunak. RAPPID Project Index Page. http://www.erim.org/cec/rappid/rappid.htm, 2002 •Weiss, Gerhard(Ed.), Parunak, H Van Dyke "Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence.", MIT Press, p378-416, 1999. ISBN •Kuokka et all, “SHADE: Knowledge-Based Technology for the Re-Engineering Problem”, http://www.ksl.stanford.edu/knowledge-sharing/papers/shadeoverview.html, 2002 •Parsons et all., “A Hybrid Agent Approach for Set-Based Conceptual Ship Design,” http://www.erim.org/cec/rappid/rappid.htm, 2002 The End