Model-Integrated Computing and Autonomous Negotiating Teams for Autonomic Logistics

1
Model-Integrated Computingand Autonomous
Negotiating Teams for Autonomic Logistics

• Gabor Karsai, Benoit Dawant
• Institute for Software-Integrated Systems,
  Vanderbilt University
• Jon Doyle, Bob Laddaga,Russ Currer
• LCS/MIT
• George Bloor,Joan Crunk, Rick Wong
• Boeing Phantom Works

2
Vision Autonomic Logisticswith Legacy Systems

• Maintenance and supply system wherein change in
  the health of aircraft triggers the logistics
  system to
• Identify, locate, gather, and schedule parts,
  equipment, and technical personnel
• Maintain stocks
• Perform data analysis and provide feedback to
  manufacturers
• Resolve conflicts and allocate scarce resources
• Constraint Utilization of existing legacy
  systems
• Potential application CACE,JSF

3
03

• Radar Altimeter Inop
• Supply Status 1 Hour

Evt 01-1/2
EVT 02-2 1300 - 1420 Capt Evans 1 CA-9, 1
Tatcs PMC/Ctr Line Sta
33
29
06
10
02
14
READY
09
07
16
05
24hrs
Today 1415
48hrs
15
72hrs
96hrs
4

• One GTS/APU on hangar deck FOM on AC undergoing
  Phase Inspection (36 hrs)
• Squadrons agree to transfer good unit and
  reassign supply request to AC in phase

• One unit on test bench AWP for a low usage diode
• ETR higher level unit 20 days
• Authorization for local purchase
• ETR 5 hours
• Total time till RFI 24hours

MAG HQTRS
Aviation Logistics Squadron
5
The MICANTS Solution

• A prototype Autonomic Logistics (AL) system using
  negotiation technology to allocate resources
• Components
• Agent-based environment for building AL systems
• Negotiation algorithms and technology
• System modeling and integration technology

6
MICANTS Concept
MIPS Environment
Models of apps, agents, etc.
Negotiating a globally beneficial solution
Models
Model Int.
Agent Space
Adapter
Adapter
Adapter
Adapter
Logistics App/Dbase (Legacy)
Logistics App/Dbase (Legacy)
Logistics App/Dbase (Legacy)
Logistics App/Dbase (Legacy)
7
ISIS Effort

1. Develop the Model-Integrated support tools for
   building the prototype systems
2. Provide a testbed for trying out novel
   negotiating algorithms and techniques
3. Realize demonstration scenarios

8
Status

• Agent framework package selected Zeus (BT)
• MIC tool development efforts
• Ontology modeling environment
• Interaction Protocol modeling environment
• Generators for synthesizing Java code (for Zeus)
• External database adapter (MS-Access,ODBC)
• Demonstration scenario and implementation

9
Ontology Modeling

• In Zeus ONTOLOGY SCHEMA
• Agents share ontologies, but not all agents need
  all ontologies
• Solution
• Global ontology models
• Agent-specific ontologies

10
Ontology Model Example
11
Interaction Protocol Modeling

• Interaction Protocol
• Sequencing of messages that constitute the
  negotiation process
• Approach
• Multiple finite-state machines with coupled
  send-receive pairs and exceptions
• Usage
• Java code is synthesized that is executed under
  Zeus

12
Interaction Protocol Model Example
13
Boeing Effort

• Demonstration scenario

14
Boeing Effort

• Model and Simulate the Decision Support Processes
  of the Marine Aircraft Group at Yuma
• Identify the utility of the MICANT technology to
  these decision support processes
• Map the MICANT negotiation technology onto these
  decision support processes

15
Status

• A Customer Has Been Identified.
• Marine Aircraft Group - 13 Yuma, Arizona
• The Domain Requirements Capture Process Has Been
  Selected.
• Model and Simulate What if engine
• The Modeling Environment Has Been Selected
• GRADE
• The Modeling and Simulation Team Has Been Formed

16
MIT Effort

• MICANTS Negotiation Approach

17
Key Concepts

• Structured change of negotiation methods
• Plans and strategies
• Goals, preferences, and utilities
• Beliefs and arguments
• Dynamic organization of negotiating parties

18
Dynamic Negotiation Strategies

• Plans specify structure of complex negotiations
• Sequential and conditional orderings
• Concurrent component activities
• Differential diagnosis and effects of situational
  changes
• Compose complex strategies from elemental methods

19
Sample Elemental Strategies

• Unpressured optimization
• Seek best deal according to goal criteria
• Sequential unpressured optimization
• Order search by participant proximity groups
• Panic mode
• Seek quickest deal, ignoring cost
• Sequential panic mode
• Shape panic offers by relation to participants

20
Autonomic Logistics Example

• Start with sequential unpressured optimization
• Ask sister squadrons, then service reserves, then
  standard suppliers, then untested suppliers
• Concurrently monitor rate of progress against
  deadlines and expectations about negotiation
  characteristics
• Transfer to sequential panic mode strategy when
  deadline nears
• Make sister squadrons best offer first, pleading
  desperation
• Use exponential bidding strategy for outside
  suppliers

21
Strategies and Goals

• Different strategies reflect different goals
• Minimizing time, personnel, facility usage,
  dollar cost
• Maximizing flexibility, robustness, readiness
• Goals concern different agents
• Narrow self-interest, group interest
• Group interest
• Shoring up weakest members
• Build up strongest members
• Sacrifice self to group goals

22
Dynamic Negotiation Goals

• Strategic progression changes goals
• Exiting information-gathering stage, entering
  hard-bargaining stage, abandon information goals
  in favor of cost-minimization goals
• Changing situation changes goals, then strategy
• Cost minimization is taking too long, give it up
  in favor of finishing quickly
• People arent taking our offers, lets change
  our cost goals
• HQ cut our budget again, lets economize
• HQ changed our mission, lets change our
  subgoals

23
Dynamic Negotiation Preferences

• Invention of preferences to cover new situations
• Bartering odd combinations of parts
• Comparing readiness for novel missions
• Toughening or liberalizing position
• Strengthen or weaken thresholds
• Add or remove factors from evaluation criteria

24
Elementary Strategic Components

• For group decisions
• Contract nets
• Market-clearing auctions
• For individual decisions
• Expected utility calculations
• Reasoned deliberation

25
Reasoned Negotiation and Deliberation

• Formulate or construct goals and preferences
  through strategy-sensitive reasoning
• Finding reasons for and against options
• Finding reasons undercutting or buttressing other
  arguments
• If utility representations required for
  efficiency, construct them from the resulting
  goals and preferences

26
Dynamic Negotiation Organization

• Relation of agent to others depends on strategy,
  situation, and history
• Construct proximity groups along different
  relational dimensions
• Shared or distinct missions
• Known or unknown quantity in negotiation history
• Authority, reliability, etc.
• Structure strategies to exploit these proximity
  groups

27
Dynamic Organization Examples

• Deal with sister squadrons, sister groups, known
  suppliers, unknown suppliers, etc.
• Resorting to unknown suppliers adds someone to
  known suppliers
• Consortia among suppliers eventuate standard
  points of contact

28
Theoretical Lessons

• Arrow impossibility theorem says any method will
  break down sometimes, unless backed up by
  dictatorial fall-back policy
• Market auctions produce optimal deals in ideal
  circumstances rare in practice

29
Practical Expectations

• Market auction approximations quickly produce
  reasonable feasibility estimates that can
  effectively guide
• Progress through negotiation plans
• Revision of negotiation goals and preferences
• Differential diagnosis between alternative
  negotiation plans

30
Demonstration scenarios
31
Structure

• MSA Maintenance Supervisor Agent
• RAA Resource Allocator Agent
• PMA Parts Manager Agent
• ESA External Supplier Agent

Website containing AVI files of demo
32
Scenario 1 Hierarchical search for suppliers

• Sequential unpressured optimization
• Round 1 with known suppliers
• PMA_x (squadron) and ESA-1 (trusted supplier)
• Round 2 (if time is available)
• ESA-2 (new supplier)

33
Scenario 2 Changing organizational structure

• ESA-2 has oversupply of parts it lowers price
• RAA monitors the deal and decides to promote
  ESA-2 to preferred supplier status

34
Scenario 3 Switching strategy function

• ESA-1 is delayed in responses
• RAA switches strategy function during the
  negotiation
• Speeds up the negotiation process but result is
  less optimal

35
Plans

• Refine scenarios with MIT, Boeing, and CACE
• Technology issues
• Enhance interaction protocol modeling
• Finish modeling environment for legacy database
  interfacing
• Investigate other agent frameworks/techniques
• Demonstration and evaluation