Autonomous Command Entities: KnowledgeBased Agents for C2 Decision Support

1
Autonomous Command Entities Knowledge-Based
Agents for C2 Decision Support

• Dr. John G. Tyler

2
Agenda

• The Problem
• Goals
• Technical Approach
• Software Agents
• Cognitive Behavior Conceptual Model
• Ontology Development
• Guide Interface
• Lessons Learned
• Next Steps

3
Problem Statement

• Cost/complexity of training Commanders/staff
• Mission applications Global Command and Control
  System (GCCS)
• Information overload
• Staff face heavy cognitive burden, extreme
  op-tempo
• Course of Action Analysis (CoAA)
• Information fusion
• Staff decision-making is inherently distributed
• Collaborative decision making environment
• Multi-echelon, multi-actor decision propagation
• Requires Increased C2 Decision Support . . .

4
C2 Decision Support
C2 Decision Support Tools to help
Commanders/staff think, plan, and decide

• Mission Rehearsal
• Dynamic Targeting

Add Decision Support capabilities for ...

• COA Generation and Analysis
• Replanning
• Wargaming

Current Generation - C2 Info Mgmt
applications (I.e., GCCS, AFATDS)
ACE
Increased Planning and Battlefield Visualization

• Collaborative Planning
• Execution Monitoring
• Map Decision Aids
• 3-D Fly Thru

Some Underlying Technologies Human Computer
Interaction Machine Learning Methods Planning and
AI-Based Design Vision and Image
Understanding Virtual Reality Alternative
Computing Paradigms Engineering of
Knowledge-Based Systems

• Dynamic Friendly Updates
• Dynamic All-Source Pictures

• Static F/E Status and Task Org.
• COA Trades
• 2-D Map Background

TODAY
Increasing Levels of C2 Decision Support
Capability
5
Goals

• Create adaptive, autonomous C2 agents
• Represent collaborative C2 decision making
  behavior
• Offload cognitive burden
• Guide operators, enhance C2 exercise
  effectiveness
• Support operational test and evaluation
• C2 Agents for test automation
• Reduce manpower requirements for executing
  scenarios
• Support training effectiveness
• C2 Agents for intelligent tutorial systems

6
Approach

• Leverage DARPA HPKB
• Inferencing tools
• Course of Action Analysis (CoAA) ontology
• Leverage USMC Conceptual Modeling effort
• Scalable, knowledge-based decision making model
• Joint Conceptual Model of the Mission Space
  (JCMMS)
• Detailed Task Analysis of USMC C2 element
• Collaborative decision-making behavior of C2
  staff
• Extend SWOP individual Marine conceptual model
• Exercise interface with legacy C2 system
• No impact to mission application code base

7
Task Domain

• Represent USMC Combat Operations Center (COC)
• Collaborative staff/operator decision making
  behavior
• Candidate C2 decision making task domains
• Course of Action analysis (CoAA - S3)
• Task organization for friendlies (Current Ops. -
  S3)
• Unit correlation for enemy forces (Intel. - S2)
• Criteria for C2 task selection
• Tractable task analysis
• Significant to Corps
• High frequency of occurrence
• ACE Focus Fire Support Coordination Center
  (FSCC)

8
Technical Approach
Software Agents
FSC Conceptual Model
Ontology Development
Guide Interface
9
Software Agents
10
Why use Agents?

• Can exploit latent processing power in a network
• Installed base of networked (IP) client PCs
• Distributed, Java-based software agents
• Available aggregate processing power gt 1K MIPS
• Available, cross-platform, low-cost development
  tools
• Commercial Off-the Shelf (COTS) tools
• Academic tools
• Use command agents to distribute C2 reasoning
• Knowledge and rules partitioned by subtask
• Communication among agents and with Cyc Server

Cyc is a registered trademark of Cycorp, Inc.
11
Desired Attributes for Interactive Agents

• Proactive - Autonomously elicits operator
  attention
• Interactive - Intuitive graphical user interface
  (GUI)
• Adaptive - Alters the agents knowledge and state
  given operator responses
• Cognitive - Represents operators dynamic mental
  model

• Metaphor
• Personal assistant
• Coach

• Capabilities
• Tentative Judgements
• Limited Guidance
• Suggestive Feedback
• Summarization

Platform
12
Benefits of Agent-Based Systems (ABS)

• Provides a distributed infrastructure for
  software system development
• Integrates heterogeneous hardware and software
  systems network-wide
• Encapsulates expertise in problem-solving
  strategy in agent for modular development
• Supports multi-agent cooperation via inter-agent
  communication language protocol
• Provides a natural model to represent
  collaboration and team problem-solving behaviors

13
Notional System Architecture
Tactical Combat Operations
AFATDS Intelligence Analysis System
MCS Intel-Ops Workstation
etc...

Div
Rgt
Agent
UNIX
UNIX
UNIX
UNIX
TCO
IAS
Operations
Intelligence
Server
Bn
NT
NT
IOW
Co
Intel./Current Ops.

ACE Agent - Pluggable C4I Decision Support
14
Development Tool JATLite

• JATLite (Java Agent Template, Lite)
• From Center for Design Research, Stanford U.
• Allows users to quickly create new software
  "agents" that communicate robustly over the
  Internet (or Intranet)
• Initial development
• A) Implement agents on multiple runtime platforms
• SGI - Irix (Unix)
• Intel - Windows 95/NT
• Sun - Solaris (Unix)
• B) Demonstrate multi-agent distributed processing
• C) Exercise intra-corporate network
  infrastructure
• http//java.stanford.edu/java_agent/html/

15
System Configuration
Guide Agent
AFATDS
FSCC Agency
World
FSC
Network
TCO
NGF
CAS
ARTY
Simulation
FSCC Knowledge Base
Cyc Engine
OTHER
Cyc Agent
16
Cognitive BehaviorConceptual Model (CM)
Fire Support Coordination - Fire Support
Execution Matrix
17
Knowledge Acquisition

• Task Focus Fire Support Coordination (FSC)
• Met with Subject Matter Experts (SMEs)
• Identified key areas of fire support coordination
• Identified areas suitable for decision support
• Consulted several doctrinal publications
• Techniques and Procedures for Fire Support
  Coordination (MCWP 3-16.2)
• Marine Corps Planning Process (MCWP 5-1)
• Targeting The Joint Targeting Process and
  Procedures for Targeting Time-Critical Targets
  (MCRP 3-16.1F)
• Marine Corps Planning Process Pocket Guide
• HPKB COA Challenge Problem Specification
• Fire Support in Combined Arms Operations

18
FSC Task Analysis

• CM documented by Formalized Data Products (FDPs)
• Structured text descriptions and supporting
  diagrams
• Created in JCMMS standard format
• FDP Meta-Data document identifies administrative
  information for an FDP
• FDP Process Description identifies processes
  performed by the FSC during fire support
  coordination
• Two FDPs created
• Fire Support Coordination
• Fire Support Execution Matrix (FSEM)
• CM correctness verified by SMEs

19
Course of Action Analysis

• A CoA must pass five criteria
• Suitability Does the COA accomplish the purpose
  and tasks and comply with commanders planning
  guidance?
• Feasibility Does the COA accomplish the mission
  within the available time, space and resources?
• Acceptability Does the COA achieve an advantage
  which justifies the cost in resources?
• Distinguishability Does the COA differ
  significantly from other COAs
• Completeness Does the COA include all tasks to
  be accomplished and describe a complete mission
  (main and supporting efforts, and associated
  risks)?

20
Constrain ScopeFire Support Coordination

• Fire Support Planning Sequence
• Mission Analysis
• Course of Action (COA) Development
• COA Analysis
• COA Comparison and Decision
• Orders Development
• Transition
• Fire Support Execution Sequence
• Battle Tracking
• Tactical Fire Direction
• Executing the Fire Support Plan
• Fire Support Replanning
• Adjust Fire Support Plan

21
Constrain ScopeCOA Development Process

• Determine Where to Find and Attack the Enemy to
  Accomplish the Essential Fire Support Tasks
  (EFSTs)
• ID High-Payoff Targets (HPTs) in Enemy
  Formations Target Value Analysis (TVA)
• Quantify Desired Effects for EFSTs
• Allocate Fire Support
• Allocate Assets to Acquire
• Allocate Assets to Attack
• Integrate Triggers with Maneuver COA
• Construct the FSEM
• Use Battle Calculus to Test Feasibility
• Assist S-2 in Collection Plan Refinement

22
Constrain ScopeConstruct FSEM Process

• Identify the HQ Issuing Orders
• List Subordinate Maneuver Units
• List Allocated Fire Support Assets
• List Target Acquisition Assets and other
  Resources
• Identify Trigger Events from Scheme of Maneuver
• Determine EFST During each Phase and Allocate
  Resources to Accomplish Them

23
FSCC Ontology Development
24
IDE-based Agent Architecture

• C runtime environment
• Integrated Development Environment runtime image
• Java runtime environment
• JATLite Java based agent development environment
• Java/IDE applications programmer interface
• From Teknowledge

25
Cycorp Tools and Languages
Cycorp
Example Axioms
(constant FireSupportControlCenter) (isa
FireSupportControlCenter
CommandOrgType) (genls FireSupportControlCenter
CommandOrg)
MELD
Commercial
Free to DOD
Tools
IDE
Cyc
subset
Languages
KE-Text
constant FireSupportControlCenter. isa
CommandOrgType. genls CommandOrg.
MELD
CycL
subset
KE Text
26
Combat Operations Center Semantic Net
Chain of Command
CO
COC
S3
S3
S2
FSC Arty CAS NGF Mtr
S2
FSC
Arty
CAS
NGF
Mtr
27
Fire Support Coordination Semantic Network
Fire Support Execution Matrix
Munitions
Weapon Systems

Targets
PL Green
HE
Pt Charlie
F-16
NSFS
CAS Sortie
ILLUM
Tank Regt 3
5/54
PL Red
WP
Arty Btn 1
F-15E
Fire Plans
CAS Sortie Target WS Munition ...
TOW
Bridge F
M198
...
...
...
28
Ontology Development

• Phase I Course of Action Analysis
• Evaluate Fire Control Plans
• Plan testing in future operations
• Determination of planbreakage in current
  operations
• Assess
• Feasibility Workable in terms of space, time,
  resources
• Suitability Accomplishes mission and adheres to
  commander guidance
• Completeness Breadth and depth
• Acceptability Risk, especially to troops

29
Ontology Development

• Phase II Fire control planning/replanning
• Plan generation
• Map assets to targets
• Work down High Payoff Target order
• Choose best available match
• Halt if either list is exhausted
• Backtrack

30
FSC Guide AgentUser Interface Design
31
Basic Guide Interface
File
View
Options
Fire Support Execution Matrix
Plan
Archer
OBJECTIVE
CONSOLID
NGF
N - Counterfire
ARTY
1 Btry
CAS
2 Sorties
2 Sorties
Messages
32
Guide InterfaceControl View Modes
33
Guide InterfaceOption Menu
34
Guide InterfaceAlert Status
File
View
Options
Fire Support Execution Matrix
Plan
Archer
OBJECTIVE
CONSOLID
NGF
N - Counterfire
ARTY
1 Btry
CAS
2 Sorties
Messages
Alert Plan break for CAS in phase CONSOLID
35
Guide InterfaceAlert Status Details
36
Guide InterfaceDisplay Alternative Plans
37
Guide InterfaceAccept Alternate Plan
38
Lessons Learned

• Supports reusable, pluggable, smart agents for
• Operational testing
• Training and Analysis
• Demonstrated potential to enhance Collaborative
  Events and Operational Assessment
• Reduce manpower requirements
• Reduce hardware requirements
• Greater coverage variability
• Free up resources to improve
• Test results analysis
• Quality and flexibility of after-action review
  (AAR) capabilities

39
Lessons Learned

• An enhanced decision-making simulation capability
• Demonstrated potential transition path to
  next-generation knowledge representation systems
• Demonstrated interoperability among relational
  databases, object-oriented databases
  ontological knowledge bases
• Demonstrated scalable solution for C2 decision
  support
• Enhanced cognitive behavior representation
• Larger, increasingly complex, aggregated entities
• Finer-grained, high-fidelity, realistic behavior
  models
• Proactive, adaptive operator interface
• Filter information
• Support/guide decision-making tasks

40
Next Steps C2 Mission App

• Interface ACE with Mission Application
• Use DARPA CoABS GRID architecture
• Build and test ACE GRID interface agent
• Exercise GCCS/ABCS messaging interface
• Build and test GRID interface agent
• Plan and Re-Plan FSEM
• Respond to battlespace events
• Southwest Asia scenarios
• Reallocate resources (platforms, munitions, etc.)
• Assure/maintain plan feasibility

41
Next StepsJSIMS OTE

• Enhance communication, collaboration
  representation
• S2 Intelligence analysis
• S3 Planning process, CoAA
• Test scenario definition and design
• E.g., USMC Amphibious Raid, Amphibious Assault
• Leverage JSIMS conceptual models
• Ontology and reasoning development
• ACE -- Automated creation of test scenario
  variations

42
Next StepsInterface Enhancements

• Alternate display views
• Example Asset Utilization
• Information visualization with 2D and 3D
  graphical displays
• Support user preferences for diverse user
  population
• Visualization enhancements
• Plan display over terrain maps
• What-If scenario comparison views

File
View
Options
Asset Usage
Stacked
Plan
Archer
Plot Options
300
200
100
CAS
NGF
ARTY
Planned Usage
Messages
Alert Plan break for CAS in phase CONSOLID
43
Your Questions?
MITRE POC Dr. John Tyler 703-883-6511
jtyler_at_mitre.org USMC POC MARCORSYSCOM
Training Systems Directorate Maj. Gary Tepera
teperags_at_mcsc.usmc.mil
44
Backup
45
Leveraging Resources
DARPA
Integrator
COA Challenge Problem
Alphatech
Teknowledge
Cycorp
GMU
MITRE Quantico
BSTF
MCOTEA
TRASYS
46
Initial Test Results

• Various platforms tested with no apparent
  difference in functionality
• 5 PCs running Win95/Win98, 1 SGI Onyx running
  IRIX 6.5 and 1 Sun running SunOS 5.5.1
• Various processors from 120Mhz to 300Mhz
• Up to 20 threads of messages producing 300,000
  messages with no network-related problems
• No noticeable impact on network resources
• Communications Integrity No messages lost

47
Inferencing Environment
Quantico Site
Eric Peterson Sun SunOS 5.5.1 Cyc
MITRE Sun SunOS 5.5.1 JATLite, Cyc
RCAgent6
Cyc, IDE
Mike Pack Dell Windows 95 JATLite Router
RCAgent5
N E T W O R K
Reston
Cyc, IDE
MITRE Sun SunOS 5.5.1 Cyc
Dave Pack Dell Windows 95 JATLite
MITRE SGI IRIX 6.5 JATLite
Mike Lee Dell Windows 95 JATLite
John Tyler Gateway 2000 Windows NT JATLite
RCAgent2
RCAgent1
RCAgent3
RCAgent4
Agent
Development
Both
Run-time test
48
Scenario Description

• Every 10 seconds the NGF and artillery agents
  query Cyc for plan breaks
• If the plan breaks, that agent will report it to
  the FSC
• The FSC will report the break to the Guide agent
• Examples of plan breaks detected
• Distance between NGF and target changes
• Weapon used by artillery team changes
• Weapon used by NGF changes
• Desired kill fraction for artillery team is
  increased
• Artillery teams target becomes dug in
• Weapon used by NGF changes
• Artillery teams target increases in size

49
Agent Roles

• Fire Support Coordinator (FSC) agent
• Coordinates activity of mission
• Periodically checks plan feasibility
• Naval Gunfire Officer (NGF) agent
• Responsible for naval gunfire aspect of mission
• Periodically checks plan feasibility
• Cyc agent
• Passes queries and assertions from agents to Cyc
• Passes results to respective agents
• World agent
• Represents what is occurring during mission
• Updates Cyc when something changes (range, ammo)

50
Evaluating Agent Development Tools

• Five factors considered in evaluation
• Agency perform tasks proactively and
  cooperatively
• Agent interpreter coordinated execution among
  agents
• Distributed infrastructure robust,
  network-centric architecture
• Programming facility ease of use, implementing
  and debugging
• Performance relatively efficient use of CPU,
  memory, etc.

51
Software Agent Toolkit Assessment
JATLite
AgentBuilder
JACK
RETSINA
Aglet
JINI
Agency
Agent Interpreter
Distributed Infrastructure
Programming Facility
Performance
52
Replanning Matrices Fire Support Degradation
Initial focus
53
Replanning Matrices Weapon System Attributes
Initial focus
54
Replanning Matrices Weapon Systems Effects
0 least likely used 5 most likely used
55
Replanning Matrices Roles and Responsibilities
FSR Arty
FSR MR
Div COC
FSR Naval
FSR AC
FSC
S2
S3
Gather info from sources
Gauge reliability of info
Determine if plan breaks
Determine plan repair options
Rank order options
Send recommendations
Re-plan if NACK
AC Aircraft COC Combat Operations Center FSC
Fire Support Coordinator FSR Fire Support
Representative MR Mortars and Rockets
56
Replanning Matrices Execution Matrix Coverage
Effect
Time
Location
Asset
0
0
0
0
1
0
0
0
1
2
0
0
1
0
3
1
1
0
0
4
0
1
0
0
5
0
1
1
0
6
1
1
0
0
7
1
1
1
0
8
0
0
1
0
9
1
1
0
0
10
1
1
0
0
11
Legend 0 unchanged 1 changed
1
1
0
1
12
1
0
1
0
13
1
1
1
0
14
0
1
1
1
Initial focus
15
1
1
1
1
16
57
Guide State Diagram
Accept User Input
Update Display
Dialog Tasks Presentation / User input
Accept/Reject Plan
Query Details
Present Plan
Present Alert
Plan Repair Tasks Compute alternatives
Alert Tasks Determine alert type Query repair
status
Feasible Plan Update
Query Repair Status
Monitor/Assess Tasks Monitor battlespace
events Assess plan impacts
Plan Break Detected
Plan Break Detected
FSC
Simulation
FSC2