Wright State University Biomedical, Industrial

1
Wright State UniversityBiomedical, Industrial
Human Factors Eng. Bay of Biscay, Agent
Modeling Study

• Raymond Hill
• Research sponsored by

2
Purpose

• Update project with DMSO/AFRL presented at last
  years conference
• AFIT Operational Sciences Department
• WSU BIE Department
• Two pieces of work accomplished to date that I
  will discuss today
• Some future plans
• Suggestions and comments?
• Sorry, I made minor changes last night

3
Quick Background on Project

• Lots of interest in agent models
• Project Albert work
• Brawler modeling work
• Next Generation Mission Model
• Other agent model work as well
• Adaptive interface agents
• Intelligent software agents
• Internet agents
• Challenge is how to bring agent models into the
  higher level models?

4
Why Higher Level Modeling?

• Need to better capture command and control
  effects
• Need to capture intangibles
• Need to model learning based on battlefield
  information
• Need better representation of actual information
  use versus perfect use
• Agents and agent models hold promise but bring
  along many issues

5
Agent Modeling Challenges

• Output analysis
• Particularly with more complex models and models
  that are not necessarily replicable
• Accurate human behavior modeling
• In particular, command behavior modeling
• Level of fidelity in model
• Beyond that of bouncing dots
• Interaction of agents and legacy modeling
  approaches
• Brawler extensions into theater and campaign
  level modeling

6
Agent Modeling Challenges (cont).

• Human interaction with the models
• The visual impact of interactions among the
  agents
• What if analyses when human behavior is being
  modeled
• Verification and Validation

7
The Project

• Need a use case for agent models
• Dr McCues book great example of operational
  analysis
• Bay of Biscay scenario amenable to agent modeling
• Lots of information available
• Forms a basis for subsequent research

8
Efforts Completed

• Capt Ron Greg Carl (masters thesis)
• Search theory focus - finished
• Capt Joe Price (masters thesis)
• Game theory focus - finished
• Subhashini Ganapathy
• Optimization study - finished
• Entering PhD candidacy
• Lance Champagne
• Dissertation defense in early Fall
• Same time twins are due!

9
Efforts Completed

• Capt Ron Greg Carl (masters thesis)
• Search theory focus - finished
• Capt Joe Price (masters thesis)
• Game theory focus - finished
• Subhashini Ganapathy
• Optimization study - finished
• Entering PhD candidacy
• Lance Champagne
• Dissertation defense in early Fall
• Same time twins are due!

10
Snapshot of AFIT Model
11
Methodology - Game Portion

• Allied search strategies
• When to search? Day versus night?
• German U-boat surfacing strategies
• When to surface? Day versus night?
• Two-person zero-sum game
• Players Allied search aircraft and German
  U-boats
• Met rationality assumption
• Non-perfect information
• Neither side knows the exact strategy the other
  uses
• Objective is number of U-boat detections
• Allied goal maximize
• German goal minimize
• Zero-sum game

12
Game Formulation

• Allies two pure search strategies
• Only day and only night
• Germans two pure surfacing strategies
• Only day and only night
• Next step to include mixed strategies
• Let parameter range from 0 to 1 as strategy
• More interesting than simple pure strategy
• Still more interesting with adaptation
• Simple adaptation algorithm
• Agents allowed to adapt strategy each month

13
Results No Adaptation

• Response Surface Methodology model
• Adjusted R2 0.947

Equilibrium Point, 0.7, 0.54
14
Adaptation Experiment

• Both sides can adapt strategies (simple model)
• Three design points chosen
• Adaptation occurs every month
• Investigate results
• 20 replications 12-month warm-up 12 months of
  statistics collection (April 1943 February 1944)

15
Adaptation Convergence
16
Adaptation Convergence
17
Methodology Search Portion

• Design data compiled according to hierarchy
• Historical fact
• Published studies
• Data derived from raw numbers
• Good judgment
• MOE is number of U-boat sightings
• U-boat density constant between replications
• Aircraft flight hours same between replications
• Therefore, sightings search efficiency
• Two cases search regions dont overlap, do
  overlap

18
Non-overlapping Search Regions
19
Overlapping Search Regions
20
Non-overlapping Search Regions
Means ComparisonAll Pairs (20 Iterations) (Simila
r Letters Indicate Statistical Equivalence)
21
Non-overlapping Search Regions
Means ComparisonAll Pairs (30 Iterations) (Simila
r Letters Indicate Statistical Equivalence)
22
Overlapping Search Regions
Means ComparisonAll Pairs (30 Iterations) (Simila
r Letters Indicate Statistical Equivalence)
23
Future Applications

• Generalized architecture promotes re-use
• Coast Guard Deep-water efforts
• Air Force UAV search in rugged terrain or urban
  environments
• Human-in-the-loop issues permeate
• Search and rescue using UAVs
• Reconnaissance using UAVs
• Combat missions using UCAVs

24
Future Efforts

• Champagne completing dissertation
• Ganapathy starting candidacy
• Looked at simulation-based optimization
• Examining human-mediated optimization techniques
• Application to search and rescue or operational
  routing
• Extensions planned
• Extend game theory aspects
• Further refinement of search results and
  optimization use

25
Questions?