Network-Aware Automated Planning and Plan Execution Kyle Usbeck A - - PowerPoint PPT Presentation

▶
network aware automated planning
SMART_READER_LITE
LIVE PREVIEW

Network-Aware Automated Planning and Plan Execution Kyle Usbeck A - - PowerPoint PPT Presentation

Dec 16, 2023 271 likes •2.04k views

Introduction Formalization Technical Approach Experiments Network-Aware Automated Planning and Plan Execution Kyle Usbeck A Thesis Submitted to the Faculty of Drexel University in partial fulfillment of the requirements for the degree of

slide-1
SLIDE 1

Computer

Department of Science

Introduction Formalization Technical Approach Experiments

Network-Aware Automated Planning

and Plan Execution Kyle Usbeck

A Thesis Submitted to the Faculty of Drexel University in partial fulfillment

  • f the requirements for the degree of

Master of Science in Computer Science

2009-07-07

Kyle Usbeck Network-Aware Automated Planning 1/75

slide-2
SLIDE 2

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 2/75

slide-3
SLIDE 3

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivation

April 2009: 75% of coalition force casualties in Afghanistan are from roadside bombs. 40% of coalition force casualties in Iraq are from roadside bombs.

Source: Tom Vanden Brook, USA Today

Kyle Usbeck Network-Aware Automated Planning 3/75

slide-4
SLIDE 4

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivation

April 2009: 75% of coalition force casualties in Afghanistan are from roadside bombs. 40% of coalition force casualties in Iraq are from roadside bombs.

Source: Tom Vanden Brook, USA Today

Kyle Usbeck Network-Aware Automated Planning 3/75

slide-5
SLIDE 5

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivating Scenario

IED Detection. Monitor Locations. Techniques. Actors. Resources. Evaluators.

location 7 location 5 location 3 location 1 location 8 location 6 location 4 location 2 robot 2 UAV 1 UAV 2 robot 1 human 1 human 3 human 2

Kyle Usbeck Network-Aware Automated Planning 4/75

slide-6
SLIDE 6

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivating Scenario

IED Detection. Monitor Locations. Techniques. Actors. Resources. Evaluators.

location 7 location 5 location 3 location 1 location 8 location 6 location 4 location 2 robot 2 UAV 1 UAV 2 robot 1 human 1 human 3 human 2

Kyle Usbeck Network-Aware Automated Planning 4/75

slide-7
SLIDE 7

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivating Scenario

IED Detection. Monitor Locations. Techniques. Actors. Resources. Evaluators.

location 7 location 5 location 3 location 1 location 8 location 6 location 4 location 2 robot 2 UAV 1 UAV 2 robot 1 human 1 human 3 human 2

Kyle Usbeck Network-Aware Automated Planning 4/75

slide-8
SLIDE 8

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivating Scenario

IED Detection. Monitor Locations. Techniques. Actors. Resources. Evaluators.

location 7 location 5 location 3 location 1 location 8 location 6 location 4 location 2 robot 2 UAV 1 UAV 2 robot 1 human 1 human 3 human 2

Kyle Usbeck Network-Aware Automated Planning 4/75

slide-9
SLIDE 9

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivating Scenario

IED Detection. Monitor Locations. Techniques. Actors. Resources. Evaluators.

location 7 location 5 location 3 location 1 location 8 location 6 location 4 location 2 robot 2 UAV 1 UAV 2 robot 1 human 1 human 3 human 2

Kyle Usbeck Network-Aware Automated Planning 4/75

slide-10
SLIDE 10

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivating Scenario

IED Detection. Monitor Locations. Techniques. Actors. Resources. Evaluators.

location 7 location 5 location 3 location 1 location 8 location 6 location 4 location 2 robot 2 UAV 1 UAV 2 robot 1 human 1 human 3 human 2

Kyle Usbeck Network-Aware Automated Planning 4/75

slide-11
SLIDE 11

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Motivating Scenario

Heterogeneous Network multiple different network technologies are combined to work together simultaneously. Network-Centric System a distributed system where performance is dependent on the quality of the underlying network communication links.

Internet MANET Wired LAN Satellite Satellite Reachback

Kyle Usbeck Network-Aware Automated Planning 5/75

slide-12
SLIDE 12

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 6/75

slide-13
SLIDE 13

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-14
SLIDE 14

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-15
SLIDE 15

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-16
SLIDE 16

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-17
SLIDE 17

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-18
SLIDE 18

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-19
SLIDE 19

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-20
SLIDE 20

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Contributions

1

Qualitatively-different plans:

Generating plans over a range of evaluation criteria; Visualizing plan evaluations. Improve plan selection.

2

Network-Aware Agents:

Classical planning domains for distributed service composition; Measuring the performance and effectiveness of planning, execution, and monitoring agents; Incorporating network-awareness.

Kyle Usbeck Network-Aware Automated Planning 7/75

slide-21
SLIDE 21

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 8/75

slide-22
SLIDE 22

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Service Composition to Automated Planning

Definition “Service composition is the linking. . . of existing services so that their aggregate behavior is that of a desired service (the goal)” [Hoffmann et al. 09]. Requires Semantic Web Services [Sirin et al. 04]. QoS Assurance [Gu et al. 03]. Assumes static networking.

Kyle Usbeck Network-Aware Automated Planning 9/75

slide-23
SLIDE 23

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Service Composition to Automated Planning

Definition “Service composition is the linking. . . of existing services so that their aggregate behavior is that of a desired service (the goal)” [Hoffmann et al. 09]. Requires Semantic Web Services [Sirin et al. 04]. QoS Assurance [Gu et al. 03]. Assumes static networking.

Kyle Usbeck Network-Aware Automated Planning 9/75

slide-24
SLIDE 24

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Service Composition to Automated Planning

Definition “Service composition is the linking. . . of existing services so that their aggregate behavior is that of a desired service (the goal)” [Hoffmann et al. 09]. Requires Semantic Web Services [Sirin et al. 04]. QoS Assurance [Gu et al. 03]. Assumes static networking.

Kyle Usbeck Network-Aware Automated Planning 9/75

slide-25
SLIDE 25

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Agents in Planning

Planner Controller

Goal(s) Model (Domain) Current State Plans

System

Actions Events

Sensor

Observations Feedback

Agents: Planning Agent. Execution Agent. Monitoring Agent. [Tate 93]

Kyle Usbeck Network-Aware Automated Planning 10/75

slide-26
SLIDE 26

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Agents in Planning

Planner Controller

Goal(s) Model (Domain) Current State Plans

System

Actions Events

Sensor

Observations Feedback

Agents: Planning Agent. Execution Agent. Monitoring Agent. [Tate 93]

Kyle Usbeck Network-Aware Automated Planning 10/75

slide-27
SLIDE 27

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Agents in Planning

Planner Controller

Goal(s) Model (Domain) Current State Plans

System

Actions Events

Sensor

Observations Feedback

Agents: Planning Agent. Execution Agent. Monitoring Agent. [Tate 93]

Kyle Usbeck Network-Aware Automated Planning 10/75

slide-28
SLIDE 28

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Planning Under Uncertainty

Restrictive Assumptions: Determinism. Full observability. Reachability goals. [Nau et al. 04] Sources of Uncertainty: Partial observability. Unreliable resources. Measurement variance. Inherently vague concepts.

Kyle Usbeck Network-Aware Automated Planning 11/75

slide-29
SLIDE 29

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Planning Under Uncertainty

Restrictive Assumptions: Determinism. Full observability. Reachability goals. [Nau et al. 04] Sources of Uncertainty: Partial observability. Unreliable resources. Measurement variance. Inherently vague concepts.

Kyle Usbeck Network-Aware Automated Planning 11/75

slide-30
SLIDE 30

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Fault Detection & Isolation (FDI)

System Residual Generation Decision Making

Types of FDI: Analytic. Data-driven. Knowledge-based. [Pettersson 05]

Kyle Usbeck Network-Aware Automated Planning 12/75

slide-31
SLIDE 31

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Fault Detection & Isolation (FDI)

Types of FDI: Analytic. Data-driven. Knowledge-based. [Pettersson 05]

Kyle Usbeck Network-Aware Automated Planning 12/75

slide-32
SLIDE 32

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Fault Detection & Isolation (FDI)

Inputs Outputs

Types of FDI: Analytic. Data-driven. Knowledge-based. [Pettersson 05]

Kyle Usbeck Network-Aware Automated Planning 12/75

slide-33
SLIDE 33

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 13/75

slide-34
SLIDE 34

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Approach

1

Modify planner to improve the quality of the plans it produces based on evaluation criteria.

2

Add network-awareness to planning, execution, and monitoring agents. Purpose To improve network-centric automated planning and execution.

Kyle Usbeck Network-Aware Automated Planning 14/75

slide-35
SLIDE 35

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Approach

1

Modify planner to improve the quality of the plans it produces based on evaluation criteria.

2

Add network-awareness to planning, execution, and monitoring agents. Purpose To improve network-centric automated planning and execution.

Kyle Usbeck Network-Aware Automated Planning 14/75

slide-36
SLIDE 36

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Motivation Background Approach

Approach

1

Modify planner to improve the quality of the plans it produces based on evaluation criteria.

2

Add network-awareness to planning, execution, and monitoring agents. Purpose To improve network-centric automated planning and execution.

Kyle Usbeck Network-Aware Automated Planning 14/75

slide-37
SLIDE 37

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 15/75

slide-38
SLIDE 38

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Σ is the planning domain — the model of the world passed as input to the planner. Σ is a Tuple S set of states; A set of actions; E set of events; γ transition function γ : S × A → S.

Kyle Usbeck Network-Aware Automated Planning 16/75

slide-39
SLIDE 39

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Σ is the planning domain — the model of the world passed as input to the planner. Σ is a Tuple S set of states; A set of actions; E set of events; γ transition function γ : S × A → S.

Kyle Usbeck Network-Aware Automated Planning 16/75

slide-40
SLIDE 40

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Σ is the planning domain — the model of the world passed as input to the planner. Σ is a Tuple S set of states; A set of actions; E set of events; γ transition function γ : S × A → S.

Kyle Usbeck Network-Aware Automated Planning 16/75

slide-41
SLIDE 41

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Σ is the planning domain — the model of the world passed as input to the planner. Σ is a Tuple S set of states; A set of actions; E set of events; γ transition function γ : S × A → S.

Kyle Usbeck Network-Aware Automated Planning 16/75

slide-42
SLIDE 42

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The functions on planning actions: For a ∈ A precond(a) preconditions of a; effects+(a) positive effects of a; effects−(a) negative effects of a; host(a) the single host h from a; resources(a) the set of resources (parameters) of action a.

Kyle Usbeck Network-Aware Automated Planning 17/75

slide-43
SLIDE 43

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The functions on planning actions: For a ∈ A precond(a) preconditions of a; effects+(a) positive effects of a; effects−(a) negative effects of a; host(a) the single host h from a; resources(a) the set of resources (parameters) of action a.

Kyle Usbeck Network-Aware Automated Planning 17/75

slide-44
SLIDE 44

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The functions on planning actions: For a ∈ A precond(a) preconditions of a; effects+(a) positive effects of a; effects−(a) negative effects of a; host(a) the single host h from a; resources(a) the set of resources (parameters) of action a.

Kyle Usbeck Network-Aware Automated Planning 17/75

slide-45
SLIDE 45

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The functions on planning actions: For a ∈ A precond(a) preconditions of a; effects+(a) positive effects of a; effects−(a) negative effects of a; host(a) the single host h from a; resources(a) the set of resources (parameters) of action a.

Kyle Usbeck Network-Aware Automated Planning 17/75

slide-46
SLIDE 46

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The functions on planning actions: For a ∈ A precond(a) preconditions of a; effects+(a) positive effects of a; effects−(a) negative effects of a; host(a) the single host h from a; resources(a) the set of resources (parameters) of action a.

Kyle Usbeck Network-Aware Automated Planning 17/75

slide-47
SLIDE 47

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The planning agent receives the tuple, IP, and creates a set of plans, PI. IP is a Tuple Σ automated planning domain; s0 initial state; Sg set of goal state(s); H set of hosts (nodes) on the network; ωH host link weighting.

Kyle Usbeck Network-Aware Automated Planning 18/75

slide-48
SLIDE 48

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The planning agent receives the tuple, IP, and creates a set of plans, PI. IP is a Tuple Σ automated planning domain; s0 initial state; Sg set of goal state(s); H set of hosts (nodes) on the network; ωH host link weighting.

Kyle Usbeck Network-Aware Automated Planning 18/75

slide-49
SLIDE 49

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The planning agent receives the tuple, IP, and creates a set of plans, PI. IP is a Tuple Σ automated planning domain; s0 initial state; Sg set of goal state(s); H set of hosts (nodes) on the network; ωH host link weighting.

Kyle Usbeck Network-Aware Automated Planning 18/75

slide-50
SLIDE 50

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The planning agent receives the tuple, IP, and creates a set of plans, PI. IP is a Tuple Σ automated planning domain; s0 initial state; Sg set of goal state(s); H set of hosts (nodes) on the network; ωH host link weighting.

Kyle Usbeck Network-Aware Automated Planning 18/75

slide-51
SLIDE 51

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

The planning agent receives the tuple, IP, and creates a set of plans, PI. IP is a Tuple Σ automated planning domain; s0 initial state; Sg set of goal state(s); H set of hosts (nodes) on the network; ωH host link weighting.

Kyle Usbeck Network-Aware Automated Planning 18/75

slide-52
SLIDE 52

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Problem To find and execute pI ∈ PI where pI = {a0, a1, . . . , a|pI|} and execution of pI yields the best domain-dependent and network-centric evaluations. Network-Awareness An agent exhibits network-awareness if changes to ωH cause the agent’s output to change while all other inputs remain constant.

Kyle Usbeck Network-Aware Automated Planning 19/75

slide-53
SLIDE 53

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Problem To find and execute pI ∈ PI where pI = {a0, a1, . . . , a|pI|} and execution of pI yields the best domain-dependent and network-centric evaluations. Network-Awareness An agent exhibits network-awareness if changes to ωH cause the agent’s output to change while all other inputs remain constant.

Kyle Usbeck Network-Aware Automated Planning 19/75

slide-54
SLIDE 54

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Host 1 Planning Agent Execution Agent Host 2 Service 1 Service 2 Host 3 Service 1 Service 3 Host 4 Monitoring Agent Service 2 Host 5 Monitoring Agent Service 3

Kyle Usbeck Network-Aware Automated Planning 20/75

slide-55
SLIDE 55

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Host 1 Planning Agent Execution Agent Plans Host 2 Host 3 Host 4 Service Calls Faults Host 5

Kyle Usbeck Network-Aware Automated Planning 21/75

slide-56
SLIDE 56

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Problem Statement

Formal Problem Statement

Planning Agent Execution Agent Monitoring Agent Plan(s) [major fault] Fault [minor fault] Fault

Kyle Usbeck Network-Aware Automated Planning 22/75

slide-57
SLIDE 57

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 23/75

slide-58
SLIDE 58

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Domain Extensions

Operator distribution e.g., NODE1ACTION(parameters) Implicit constraints. Resource distribution e.g., ACTION(node1, parameters) s0 ← s0 ∪ {TYPE(node1) = NETWORKNODE} s0 ← s0 ∪ {ACTION(node1) = true}

Kyle Usbeck Network-Aware Automated Planning 24/75

slide-59
SLIDE 59

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Domain Extensions

Operator distribution e.g., NODE1ACTION(parameters) Implicit constraints. Resource distribution e.g., ACTION(node1, parameters) s0 ← s0 ∪ {TYPE(node1) = NETWORKNODE} s0 ← s0 ∪ {ACTION(node1) = true}

Kyle Usbeck Network-Aware Automated Planning 24/75

slide-60
SLIDE 60

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Domain Extensions

Operator distribution e.g., NODE1ACTION(parameters) Implicit constraints. Resource distribution e.g., ACTION(node1, parameters) s0 ← s0 ∪ {TYPE(node1) = NETWORKNODE} s0 ← s0 ∪ {ACTION(node1) = true}

Kyle Usbeck Network-Aware Automated Planning 24/75

slide-61
SLIDE 61

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Domain Extensions

Operator distribution e.g., NODE1ACTION(parameters) Implicit constraints. Resource distribution e.g., ACTION(node1, parameters) s0 ← s0 ∪ {TYPE(node1) = NETWORKNODE} s0 ← s0 ∪ {ACTION(node1) = true}

Kyle Usbeck Network-Aware Automated Planning 24/75

slide-62
SLIDE 62

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Domain Extensions

Operator distribution e.g., NODE1ACTION(parameters) Implicit constraints. Resource distribution e.g., ACTION(node1, parameters) s0 ← s0 ∪ {TYPE(node1) = NETWORKNODE} s0 ← s0 ∪ {ACTION(node1) = true}

Kyle Usbeck Network-Aware Automated Planning 24/75

slide-63
SLIDE 63

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Domain Extensions

Operator distribution e.g., NODE1ACTION(parameters) Implicit constraints. Resource distribution e.g., ACTION(node1, parameters) s0 ← s0 ∪ {TYPE(node1) = NETWORKNODE} s0 ← s0 ∪ {ACTION(node1) = true} Complexity Operator distribution increases the number of actions in Σ to |H| × |A| in the worst case. Resource distributed increases the number of constraints in the world-state.

Kyle Usbeck Network-Aware Automated Planning 24/75

slide-64
SLIDE 64

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Agents

Agent Types: Domain-Independent. Random. Guided. Plan Evaluators: Steps. Alternatives. Longest temporally

  • rdered path.

Duplicate plans.

Kyle Usbeck Network-Aware Automated Planning 25/75

slide-65
SLIDE 65

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Agents

Agent Types: Domain-Independent. Random. Guided. Plan Evaluators: (none).

Kyle Usbeck Network-Aware Automated Planning 25/75

slide-66
SLIDE 66

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Planning Agents

Agent Types: Domain-Independent. Random. Guided. Plan Evaluators: IED detection accuracy. Plan execution time. Network link quality. Network bandwidth usage.

Kyle Usbeck Network-Aware Automated Planning 25/75

slide-67
SLIDE 67

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Domain-Independent Planning Agent

Uses I-Plan’s default strategy. I-Plan University of Edinburgh, Tate et al. ’s plan-space HTN planner which is built on an intelligent agent framework, I-X. Process

1

Traverses search space depth-first.

2

Encounter an alternative whose constraints cannot be satisfied.

3

Backtracks using an A* search.

Kyle Usbeck Network-Aware Automated Planning 26/75

slide-68
SLIDE 68

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Domain-Independent Planning Agent

Uses I-Plan’s default strategy. I-Plan University of Edinburgh, Tate et al. ’s plan-space HTN planner which is built on an intelligent agent framework, I-X. Process

1

Traverses search space depth-first.

2

Encounter an alternative whose constraints cannot be satisfied.

3

Backtracks using an A* search.

Kyle Usbeck Network-Aware Automated Planning 26/75

slide-69
SLIDE 69

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Domain-Independent Planning Agent

Uses I-Plan’s default strategy. I-Plan University of Edinburgh, Tate et al. ’s plan-space HTN planner which is built on an intelligent agent framework, I-X. Process

1

Traverses search space depth-first.

2

Encounter an alternative whose constraints cannot be satisfied.

3

Backtracks using an A* search.

Kyle Usbeck Network-Aware Automated Planning 26/75

slide-70
SLIDE 70

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Domain-Independent Planning Agent

Uses I-Plan’s default strategy. I-Plan University of Edinburgh, Tate et al. ’s plan-space HTN planner which is built on an intelligent agent framework, I-X. Process

1

Traverses search space depth-first.

2

Encounter an alternative whose constraints cannot be satisfied.

3

Backtracks using an A* search.

Kyle Usbeck Network-Aware Automated Planning 26/75

slide-71
SLIDE 71

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Domain-Independent Planning Agent

Uses I-Plan’s default strategy. I-Plan University of Edinburgh, Tate et al. ’s plan-space HTN planner which is built on an intelligent agent framework, I-X. Process

1

Traverses search space depth-first.

2

Encounter an alternative whose constraints cannot be satisfied.

3

Backtracks using an A* search.

Kyle Usbeck Network-Aware Automated Planning 26/75

slide-72
SLIDE 72

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Random Planning Agent

DFS with random branching. Process CONSTRUCTRANDOMPLAN(IP)

1: toVisit.push(s0) 2: while ¬ toVisit.empty() ∧ ¬ solution(toVisit.peek()) do 3:

v ← toVisit.pop()

4:

if v / ∈ visited then

5:

visited.add(v)

6:

r ← randomize(v.children())

7:

toVisit.push(r)

8:

end if

9: end while 10: return toVisit.peek()

Kyle Usbeck Network-Aware Automated Planning 27/75

slide-73
SLIDE 73

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Guided Planning Agent

Generates qualitatively-different plans over: Domain-dependent criteria, and Network-centric criteria. Process

1

A priority queue exists for each evaluator.

2

Every partial-plan is evaluated by all evaluators and placed in their respective priority queues.

3

The partial-plan at the head of each priority queue is used for the next step.

Kyle Usbeck Network-Aware Automated Planning 28/75

slide-74
SLIDE 74

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Guided Planning Agent

Generates qualitatively-different plans over: Domain-dependent criteria, and Network-centric criteria. Process

1

A priority queue exists for each evaluator.

2

Every partial-plan is evaluated by all evaluators and placed in their respective priority queues.

3

The partial-plan at the head of each priority queue is used for the next step.

Kyle Usbeck Network-Aware Automated Planning 28/75

slide-75
SLIDE 75

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Guided Planning Agent

Generates qualitatively-different plans over: Domain-dependent criteria, and Network-centric criteria. Process

1

A priority queue exists for each evaluator.

2

Every partial-plan is evaluated by all evaluators and placed in their respective priority queues.

3

The partial-plan at the head of each priority queue is used for the next step.

Kyle Usbeck Network-Aware Automated Planning 28/75

slide-76
SLIDE 76

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Guided Planning Agent

Generates qualitatively-different plans over: Domain-dependent criteria, and Network-centric criteria. Process

1

A priority queue exists for each evaluator.

2

Every partial-plan is evaluated by all evaluators and placed in their respective priority queues.

3

The partial-plan at the head of each priority queue is used for the next step.

Kyle Usbeck Network-Aware Automated Planning 28/75

slide-77
SLIDE 77

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Guided Planning Agent

Planner new partial-plan Evaluator 1 Priority Queue Evaluator 2 Priority Queue Evaluator 3 Priority Queue Evaluator 4 Priority Queue Kyle Usbeck Network-Aware Automated Planning 29/75

slide-78
SLIDE 78

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Guided Planning Agent

Planner Evaluator 1 Priority Queue Evaluator 2 Priority Queue Evaluator 3 Priority Queue Evaluator 4 Priority Queue evaluations next partial-plan Kyle Usbeck Network-Aware Automated Planning 30/75

slide-79
SLIDE 79

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Guided Planning Agent

Planner new partial-plan Evaluator 1 Priority Queue Evaluator 2 Priority Queue Evaluator 3 Priority Queue Evaluator 4 Priority Queue Kyle Usbeck Network-Aware Automated Planning 31/75

slide-80
SLIDE 80

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 32/75

slide-81
SLIDE 81

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Execution Agents

Execution Agent pI Service calls Faults PI Plan Selection

Agent types: Naïve. Reactive. Proactive. Defined by: Service invocation. Error handling.

Kyle Usbeck Network-Aware Automated Planning 33/75

slide-82
SLIDE 82

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Execution Agents

Execution Agent pI Service calls Faults PI Plan Selection

Agent types: Naïve. Reactive. Proactive. Defined by: Service invocation. Error handling.

Kyle Usbeck Network-Aware Automated Planning 33/75

slide-83
SLIDE 83

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Naïve Execution Agent

Naïve Execution Agent Properties Service Invocation Invokes services exactly as described by pI. The naïve agent requires that ∀ actions a ∈ pI, host(a) = ∅ ∧ resources(a) = {}. Error Handling Ignores execution errors. Not network-aware.

Kyle Usbeck Network-Aware Automated Planning 34/75

slide-84
SLIDE 84

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Naïve Execution Agent

Naïve Execution Agent Properties Service Invocation Invokes services exactly as described by pI. The naïve agent requires that ∀ actions a ∈ pI, host(a) = ∅ ∧ resources(a) = {}. Error Handling Ignores execution errors. Not network-aware.

Kyle Usbeck Network-Aware Automated Planning 34/75

slide-85
SLIDE 85

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Naïve Execution Agent

Naïve Execution Agent Properties Service Invocation Invokes services exactly as described by pI. The naïve agent requires that ∀ actions a ∈ pI, host(a) = ∅ ∧ resources(a) = {}. Error Handling Ignores execution errors. Not network-aware.

Kyle Usbeck Network-Aware Automated Planning 34/75

slide-86
SLIDE 86

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Reactive Execution Agent

Reactive Execution Agent Properties Service Invocation Invokes services exactly as described by pI. The reactive agent requires that ∀ actions a ∈ pI, host(a) = ∅ ∧ resources(a) = {}. Error Handling Repairs the failed pI by replacing failed service call(s) with new ones, creating p′

I.

Network-aware recovery — plan repair. Uses routing protocol neighbors & link quality.

Kyle Usbeck Network-Aware Automated Planning 35/75

slide-87
SLIDE 87

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Reactive Execution Agent

Reactive Execution Agent Properties Service Invocation Invokes services exactly as described by pI. The reactive agent requires that ∀ actions a ∈ pI, host(a) = ∅ ∧ resources(a) = {}. Error Handling Repairs the failed pI by replacing failed service call(s) with new ones, creating p′

I.

Network-aware recovery — plan repair. Uses routing protocol neighbors & link quality.

Kyle Usbeck Network-Aware Automated Planning 35/75

slide-88
SLIDE 88

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Reactive Execution Agent

Reactive Execution Agent Properties Service Invocation Invokes services exactly as described by pI. The reactive agent requires that ∀ actions a ∈ pI, host(a) = ∅ ∧ resources(a) = {}. Error Handling Repairs the failed pI by replacing failed service call(s) with new ones, creating p′

I.

Network-aware recovery — plan repair. Uses routing protocol neighbors & link quality.

Kyle Usbeck Network-Aware Automated Planning 35/75

slide-89
SLIDE 89

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Reactive Execution Agent

Reactive Execution Agent Properties Service Invocation Invokes services exactly as described by pI. The reactive agent requires that ∀ actions a ∈ pI, host(a) = ∅ ∧ resources(a) = {}. Error Handling Repairs the failed pI by replacing failed service call(s) with new ones, creating p′

I.

Network-aware recovery — plan repair. Uses routing protocol neighbors & link quality.

Kyle Usbeck Network-Aware Automated Planning 35/75

slide-90
SLIDE 90

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Reactive Execution Agent

has more actions?

yes

Start Accept Plan failed?

no yes

repaired?

yes

Failure

no

Execute Next Action Repair Plan Success

no Kyle Usbeck Network-Aware Automated Planning 36/75

slide-91
SLIDE 91

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Proactive Execution Agent

Proactive Execution Agent Properties Service Invocation Invokes services using network-aware logic to choose the host and resources at execution

  • time. The proactive execution agent uses only

service descriptions from actions a ∈ pI, meaning ∀a ∈ pI, host(a) = ∅ ∧ resources(a) = {} Error Handling Repairs the failed pI by replacing failed service call(s) with new ones, creating p′

I.

Network-aware host/resource grounding.

Kyle Usbeck Network-Aware Automated Planning 37/75

slide-92
SLIDE 92

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Proactive Execution Agent

Proactive Execution Agent Properties Service Invocation Invokes services using network-aware logic to choose the host and resources at execution

  • time. The proactive execution agent uses only

service descriptions from actions a ∈ pI, meaning ∀a ∈ pI, host(a) = ∅ ∧ resources(a) = {} Error Handling Repairs the failed pI by replacing failed service call(s) with new ones, creating p′

I.

Network-aware host/resource grounding.

Kyle Usbeck Network-Aware Automated Planning 37/75

slide-93
SLIDE 93

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Proactive Execution Agent

Proactive Execution Agent Properties Service Invocation Invokes services using network-aware logic to choose the host and resources at execution

  • time. The proactive execution agent uses only

service descriptions from actions a ∈ pI, meaning ∀a ∈ pI, host(a) = ∅ ∧ resources(a) = {} Error Handling Repairs the failed pI by replacing failed service call(s) with new ones, creating p′

I.

Network-aware host/resource grounding.

Kyle Usbeck Network-Aware Automated Planning 37/75

slide-94
SLIDE 94

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Proactive Execution Agent

has more actions?

yes

Start Accept Unground Plan failed?

no yes

repaired?

yes

Failure

no

Ground First Plan Action Repair Plan Success

no

Execute Ground Plan Action

Kyle Usbeck Network-Aware Automated Planning 38/75

slide-95
SLIDE 95

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 39/75

slide-96
SLIDE 96

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Monitoring Agents

Methods of FDI

1

Analytic.

2

Data-driven.

3

Knowledge-based.

Kyle Usbeck Network-Aware Automated Planning 40/75

slide-97
SLIDE 97

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Monitoring Agents

Methods of FDI

1

  • Analytic. ← Active Monitor

2

Data-driven. ← Passive Monitor

3

Knowledge-based.

Kyle Usbeck Network-Aware Automated Planning 40/75

slide-98
SLIDE 98

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Analytic Monitoring Agent

Given the ordered plan pI = {a0, a1, . . . , a|pI|} An analytic monitoring agent:

1

Constructs pM = {m0, m1, . . . , m|pI|+1}, an ordered set of monitoring actions;

2

Creates the new execution plan p′

I = n i=0{mi, ai};

3

The result is p′

I = {m0, a0, m1, a1, . . . , m|pI|, a|pI|, m|pI|+1}.

4

Each m ∈ pM calculates the residual between expected and actual bytes transferred.

Kyle Usbeck Network-Aware Automated Planning 41/75

slide-99
SLIDE 99

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Analytic Monitoring Agent

Given the ordered plan pI = {a0, a1, . . . , a|pI|} An analytic monitoring agent:

1

Constructs pM = {m0, m1, . . . , m|pI|+1}, an ordered set of monitoring actions;

2

Creates the new execution plan p′

I = n i=0{mi, ai};

3

The result is p′

I = {m0, a0, m1, a1, . . . , m|pI|, a|pI|, m|pI|+1}.

4

Each m ∈ pM calculates the residual between expected and actual bytes transferred.

Kyle Usbeck Network-Aware Automated Planning 41/75

slide-100
SLIDE 100

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Analytic Monitoring Agent

Given the ordered plan pI = {a0, a1, . . . , a|pI|} An analytic monitoring agent:

1

Constructs pM = {m0, m1, . . . , m|pI|+1}, an ordered set of monitoring actions;

2

Creates the new execution plan p′

I = n i=0{mi, ai};

3

The result is p′

I = {m0, a0, m1, a1, . . . , m|pI|, a|pI|, m|pI|+1}.

4

Each m ∈ pM calculates the residual between expected and actual bytes transferred.

Kyle Usbeck Network-Aware Automated Planning 41/75

slide-101
SLIDE 101

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Analytic Monitoring Agent

Given the ordered plan pI = {a0, a1, . . . , a|pI|} An analytic monitoring agent:

1

Constructs pM = {m0, m1, . . . , m|pI|+1}, an ordered set of monitoring actions;

2

Creates the new execution plan p′

I = n i=0{mi, ai};

3

The result is p′

I = {m0, a0, m1, a1, . . . , m|pI|, a|pI|, m|pI|+1}.

4

Each m ∈ pM calculates the residual between expected and actual bytes transferred.

Kyle Usbeck Network-Aware Automated Planning 41/75

slide-102
SLIDE 102

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Analytic Monitoring Agent

Given the ordered plan pI = {a0, a1, . . . , a|pI|} An analytic monitoring agent:

1

Constructs pM = {m0, m1, . . . , m|pI|+1}, an ordered set of monitoring actions;

2

Creates the new execution plan p′

I = n i=0{mi, ai};

3

The result is p′

I = {m0, a0, m1, a1, . . . , m|pI|, a|pI|, m|pI|+1}.

4

Each m ∈ pM calculates the residual between expected and actual bytes transferred.

Kyle Usbeck Network-Aware Automated Planning 41/75

slide-103
SLIDE 103

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Analytic Monitoring Agent

Execution Agent Monitoring Agent 1 Starting Action Action Execution Ending Action

...

Monitoring Agent 1 Residual Monitoring Agent n

... ...

Monitoring Agent n Residual Fault Detection

Kyle Usbeck Network-Aware Automated Planning 42/75

slide-104
SLIDE 104

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Data-driven Monitoring Agent

Execution Agent Data-driven Monitoring Agent 1 Data-driven Monitoring Agent 2 Data-driven Monitoring Agent n

...

Faults Kyle Usbeck Network-Aware Automated Planning 43/75

slide-105
SLIDE 105

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Data-driven Monitoring Agent

1 10 100 1000 10000 100000 10 20 30 40 50 60 70 (See legend for units) Elapsed Time (sec) Network Statistics During Host Disconnection Packets Sent Data Packets Retransmitted Data Packets Received Packets TCP ACKs Duplicate TCP ACKs Completely Duplicate Packets Old Duplicate Packets Window Update Packets Segments Updated RTT Retransmission Timeouts

Multivariate monitor. Data packets. Retransmission timeouts.

Kyle Usbeck Network-Aware Automated Planning 44/75

slide-106
SLIDE 106

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Data-driven Monitoring Agent

1 10 100 1000 10000 100000 10 20 30 40 50 60 70 (See legend for units) Elapsed Time (sec) Network Statistics During Host Disconnection Packets Sent Data Packets Retransmitted Data Packets Received Packets TCP ACKs Duplicate TCP ACKs Completely Duplicate Packets Old Duplicate Packets Window Update Packets Segments Updated RTT Retransmission Timeouts

Multivariate monitor. Data packets. Retransmission timeouts.

Kyle Usbeck Network-Aware Automated Planning 44/75

slide-107
SLIDE 107

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 45/75

slide-108
SLIDE 108

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Plan Evaluation Criteria Statistics

Aspects Range (effective and theoretic). Direction (minimize or maximize). Statistics (e.g., mean, median, mode, standard deviation). Benefit Plans can be positioned along an absolute continuum of evaluation values.

Kyle Usbeck Network-Aware Automated Planning 46/75

slide-109
SLIDE 109

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Plan Evaluation Criteria Statistics

Aspects Range (effective and theoretic). Direction (minimize or maximize). Statistics (e.g., mean, median, mode, standard deviation). Benefit Plans can be positioned along an absolute continuum of evaluation values.

Kyle Usbeck Network-Aware Automated Planning 46/75

slide-110
SLIDE 110

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Plan Evaluation Criteria Statistics

Aspects Range (effective and theoretic). Direction (minimize or maximize). Statistics (e.g., mean, median, mode, standard deviation). Benefit Plans can be positioned along an absolute continuum of evaluation values.

Kyle Usbeck Network-Aware Automated Planning 46/75

slide-111
SLIDE 111

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Plan Evaluation Criteria Statistics

Aspects Range (effective and theoretic). Direction (minimize or maximize). Statistics (e.g., mean, median, mode, standard deviation). Benefit Plans can be positioned along an absolute continuum of evaluation values.

Kyle Usbeck Network-Aware Automated Planning 46/75

slide-112
SLIDE 112

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Dominant Plans

Definition A plan, p, is dominant to a set of other plans, P− in respect to two or more plan evaluators e1...k ∈ E when ∀e ∈ E, p− ∈ P−[e(p) ≥ e(p−)].

Plan Evaluation 1 Plan Evaluation 2

Kyle Usbeck Network-Aware Automated Planning 47/75

slide-113
SLIDE 113

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI

Plan Evaluation Visualization

Kyle Usbeck Network-Aware Automated Planning 48/75

slide-114
SLIDE 114

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 49/75

slide-115
SLIDE 115

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Experiment: Plan Evaluation Benchmarking

Location 1 Location 4 Location 3 Location 2 Camera 1 Node 1 Node 2 Node 3 Node 4 Node 5 Camera 2

Kyle Usbeck Network-Aware Automated Planning 50/75

slide-116
SLIDE 116

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Evaluation Benchmarking

Action Providing Hosts

PHYSICALMOVE

all

ACQUIRECAMERA

all

TAKEPHOTO

all

GETOLDPHOTO

all

RELEASECAMERA

all

CHECKFORIEDAT

1, 2, and 5

MANUALSEARCH

1, 2, 3, and 4

PHOTOGRAPHICSEARCH

3, 4, and 5

PHOTOARCHIVE

5

PHOTOCOMPARE

4 and 5

RESULTREPORT

2 and 5

Kyle Usbeck Network-Aware Automated Planning 51/75

slide-117
SLIDE 117

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Evaluation Benchmarking

Camera Resolution Camera 1 3.2 MP Camera 2 8.0 MP Node Speed (max mph) Transportation Cost ($ per mile) Node 1 30 6.0 Node 2 40 6.5 Node 3 20 5.1 Node 4 10 4.9 Node 5 45 6.2

Kyle Usbeck Network-Aware Automated Planning 52/75

slide-118
SLIDE 118

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Evaluation Benchmarking Results

Each planning algorithm ran in I-Plan for five minutes. σ Plan Evaluations ωH Bandwidth IED Acc. Time I-Plan Default 0.949 0.759 291.4 8216 Random 1.647 1.476 177.9 7220 Guided 1.916 1.141 392.6 14050 Dominant Plans Search Strategy % Dominant Plans Produced I-Plan Default 7.4% Random 33.3% Guided 59.3%

Kyle Usbeck Network-Aware Automated Planning 53/75

slide-119
SLIDE 119

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Evaluation Benchmarking Results

Each planning algorithm ran in I-Plan for five minutes. σ Plan Evaluations ωH Bandwidth IED Acc. Time I-Plan Default 0.949 0.759 291.4 8216 Random 1.647 1.476 177.9 7220 Guided 1.916 1.141 392.6 14050 Dominant Plans Search Strategy % Dominant Plans Produced I-Plan Default 7.4% Random 33.3% Guided 59.3%

Kyle Usbeck Network-Aware Automated Planning 53/75

slide-120
SLIDE 120

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 54/75

slide-121
SLIDE 121

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Experiment: Network-Aware Agent Combinations

Agent Technique Planning Random Domain-independent (I-Plan) Guided Execution Naïve Reactive Proactive Monitoring Data-driven Analytic (none)

Kyle Usbeck Network-Aware Automated Planning 55/75

slide-122
SLIDE 122

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Experimental Setup

Multi-objective Optimization (MOO) Function. Implemented agents with I-X and I-Plan. Network emulation. Mobility models. MOO function MOO(pI) = IEDDetectAcc(pI) + 3 × TranspCost(pI) + 5 × ExecTime(pI) + LinkQuality(pI) + BandwidthUse(pI)

Kyle Usbeck Network-Aware Automated Planning 56/75

slide-123
SLIDE 123

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Experimental Setup

Multi-objective Optimization (MOO) Function. Implemented agents with I-X and I-Plan. Network emulation. Mobility models. MOO function MOO(pI) = IEDDetectAcc(pI) + 3 × TranspCost(pI) + 5 × ExecTime(pI) + LinkQuality(pI) + BandwidthUse(pI)

Kyle Usbeck Network-Aware Automated Planning 56/75

slide-124
SLIDE 124

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Experimental Setup

Multi-objective Optimization (MOO) Function. Implemented agents with I-X and I-Plan. Network emulation. Mobility models. MOO function MOO(pI) = IEDDetectAcc(pI) + 3 × TranspCost(pI) + 5 × ExecTime(pI) + LinkQuality(pI) + BandwidthUse(pI)

Kyle Usbeck Network-Aware Automated Planning 56/75

slide-125
SLIDE 125

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Experimental Setup

Multi-objective Optimization (MOO) Function. Implemented agents with I-X and I-Plan. Network emulation. Mobility models. MOO function MOO(pI) = IEDDetectAcc(pI) + 3 × TranspCost(pI) + 5 × ExecTime(pI) + LinkQuality(pI) + BandwidthUse(pI)

Kyle Usbeck Network-Aware Automated Planning 56/75

slide-126
SLIDE 126

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

CORE

Boeing’s Common Open Research Emulator. FreeBSD network stack emulation. Simple Multicast Forwarding (SMF). Open Shortest Path First (OSPF).

Kyle Usbeck Network-Aware Automated Planning 57/75

slide-127
SLIDE 127

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

CORE

Boeing’s Common Open Research Emulator. FreeBSD network stack emulation. Simple Multicast Forwarding (SMF). Open Shortest Path First (OSPF).

Kyle Usbeck Network-Aware Automated Planning 57/75

slide-128
SLIDE 128

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

CORE

Boeing’s Common Open Research Emulator. FreeBSD network stack emulation. Simple Multicast Forwarding (SMF). Open Shortest Path First (OSPF).

Kyle Usbeck Network-Aware Automated Planning 57/75

slide-129
SLIDE 129

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

CORE

Boeing’s Common Open Research Emulator. FreeBSD network stack emulation. Simple Multicast Forwarding (SMF). Open Shortest Path First (OSPF).

Kyle Usbeck Network-Aware Automated Planning 57/75

slide-130
SLIDE 130

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-131
SLIDE 131

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-132
SLIDE 132

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-133
SLIDE 133

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-134
SLIDE 134

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Location 1 Location 4 Location 3 Location 2 Camera 1 Node 1 Node 2 Node 3 Node 4 Node 5 Camera 2

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-135
SLIDE 135

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-136
SLIDE 136

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-137
SLIDE 137

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Mobility Models

Purpose Dictate geographical node locations. Dynamic ωH. Mobility Patterns

1

Local.

2

Static.

3

Dynamic.

4

Partition-merge.

Kyle Usbeck Network-Aware Automated Planning 58/75

slide-138
SLIDE 138

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Domain-independent Plan

checkForIEDAt location1 manualSearch node1 location1 physicalMove node1 location1 conductScan node1 location1 physicalMove node2 location1 reportResults node2 location1 checkForIEDAt location2 manualSearch node1 location2 physicalMove node1 location2 conductScan node1 location2 physicalMove node2 location2 reportResults node2 location2 checkForIEDAt location3 manualSearch node1 location3 physicalMove node1 location3 conductScan node1 location3 physicalMove node2 location3 reportResults node2 location3 Kyle Usbeck Network-Aware Automated Planning 59/75

slide-139
SLIDE 139

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Random Plan

checkForIEDAt location1 photographicSearch node3 location1 physicalMoveToCamera node3 camera1 acquireCamera node3 location1 camera1 physicalMove node3 location1 getOldPhoto node5 to photo-0 takePhoto node3 location1 camera1 to photo-1 comparePhotos node4 photo-1 photo-0 reportResults node2 location1 checkForIEDAt location2 manualSearch node1 location2 physicalMove node1 location2 conductScan node1 location2 physicalMove node2 location2 reportResults node2 location2 checkForIEDAt location3 manualSearch node1 location3 physicalMove node1 location3 conductScan node1 location3 physicalMove node2 location3 reportResults node2 location3 Kyle Usbeck Network-Aware Automated Planning 60/75

slide-140
SLIDE 140

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Guided Plan

checkForIEDAt location1 photographicSearch node5 location1 physicalMoveToCamera node5 camera2 acquireCamera node5 location1 camera2 physicalMove node5 location1 getOldPhoto node5 to photo-0 takePhoto node5 location1 camera2 to photo-1 comparePhotos node5 photo-1 photo-0 reportResults node5 location1 checkForIEDAt location2 manualSearch node3 location2 physicalMove node3 location2 conductScan node3 location2 physicalMove node5 location2 reportResults node5 location2 checkForIEDAt location3 manualSearch node4 location3 physicalMove node4 location3 conductScan node4 location3 physicalMove node2 location3 reportResults node2 location3 Kyle Usbeck Network-Aware Automated Planning 61/75

slide-141
SLIDE 141

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Local Results: Mean Time

2 4 6 8 10 Naive Reactive (analytic) Reactive (data-driven) Execution Time (min) Execution Agent Planning Agent I-Plan Random Guided

Network not a factor. Network-awareness did not hurt.

Kyle Usbeck Network-Aware Automated Planning 62/75

slide-142
SLIDE 142

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Local Results: Mean IED Detection Accuracy

20 40 60 80 100 I-Plan Random Guided IED Detection Accuracy (%) Planning Agent

Ideal values of IED detection accuracy.

Kyle Usbeck Network-Aware Automated Planning 63/75

slide-143
SLIDE 143

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Planning Agent Comparison

2 4 6 8 10 I-Plan Random Guided Execution Time (min) Planning Agent Network Dynamism Static Dynamic Part-Merge

Network disruptions adversely effect plan execution times. Guided was 16.7% faster than I-Plan and 28.8% faster than random in part-merge.

Kyle Usbeck Network-Aware Automated Planning 64/75

slide-144
SLIDE 144

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Execution Agent Effectiveness

Planning Agent: domain-independent (I-Plan default)

2 4 6 8 10 20 40 60 80 100 Plan execution time (min) IED detection accuracy (%) Naive Reactive Proactive

Naïve agent has the lowest IED detection accuracy and exec. time. Reactive and proactive agents achieved ideal IED detection accuracies.

Kyle Usbeck Network-Aware Automated Planning 65/75

slide-145
SLIDE 145

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Execution Agent Effectiveness

Planning Agent: random

2 4 6 8 10 20 40 60 80 100 Plan execution time (min) IED detection accuracy (%) Naive Reactive Proactive

Naïve agent failed most often. Proactive agent finished considerably faster than reactive.

Kyle Usbeck Network-Aware Automated Planning 66/75

slide-146
SLIDE 146

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Execution Agent Effectiveness

Planning Agent: guided (network-aware)

2 4 6 8 10 20 40 60 80 100 Plan execution time (min) IED detection accuracy (%) Naive Reactive Proactive

Naïve agent failed most often. The guided algorithm advice significantly helped the execution agent.

Kyle Usbeck Network-Aware Automated Planning 67/75

slide-147
SLIDE 147

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Execution Agent Performance

20 40 60 80 100 120 140 Naive Reactive Proactive Total number of packets transferred Execution Agent Network Dynamism Static Dynamic Part-Merge

Proactive agent uses slightly more network transmissions under connected mobility patterns. Under part-merge, the proactive agent sent fewer than half as many packets as the reactive agent.

Kyle Usbeck Network-Aware Automated Planning 68/75

slide-148
SLIDE 148

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Monitoring Agent Comparisons

Analytic Monitoring Agent High percentage of false-positives. Communication errors → incorrect residuals. Active monitor. Analytic monitors are less-suitable for network-centric domains.

Kyle Usbeck Network-Aware Automated Planning 69/75

slide-149
SLIDE 149

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Monitoring Agent Comparisons

Analytic Monitoring Agent High percentage of false-positives. Communication errors → incorrect residuals. Active monitor. Analytic monitors are less-suitable for network-centric domains.

Kyle Usbeck Network-Aware Automated Planning 69/75

slide-150
SLIDE 150

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Monitoring Agent Comparisons

Analytic Monitoring Agent High percentage of false-positives. Communication errors → incorrect residuals. Active monitor. Analytic monitors are less-suitable for network-centric domains.

Kyle Usbeck Network-Aware Automated Planning 69/75

slide-151
SLIDE 151

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Monitoring Agent Comparisons

Analytic Monitoring Agent High percentage of false-positives. Communication errors → incorrect residuals. Active monitor. Analytic monitors are less-suitable for network-centric domains.

Kyle Usbeck Network-Aware Automated Planning 69/75

slide-152
SLIDE 152

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Data-driven Monitoring Agent

Normal execution:

20 40 60 80 100 120 140 160 50 100 150 200 250 1 2 3 4 5 6 7 Total number of data packets processed Total number of TCP retransmit timeouts Elapsed Time (sec) Data-driven Monitoring Agent During Dynamic Link Mobility Number of data packets TCP Retransmit Timeouts

Kyle Usbeck Network-Aware Automated Planning 70/75

slide-153
SLIDE 153

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Data-driven Monitoring Agent

Network disconnection:

20 40 60 80 100 120 140 160 50 100 150 200 250 300 350 1 2 3 4 5 6 7 Total number of data packets processed Total number of TCP retransmit timeouts Elapsed Time (sec) Data-driven Monitoring Agent During Partition/Merge Mobility Number of data packets TCP Retransmit Timeouts

Kyle Usbeck Network-Aware Automated Planning 71/75

slide-154
SLIDE 154

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Data-driven Monitoring Agent

Network disconnection:

20 40 60 80 100 120 140 160 50 100 150 200 250 300 350 1 2 3 4 5 6 7 Total number of data packets processed Total number of TCP retransmit timeouts Elapsed Time (sec) Data-driven Monitoring Agent During Partition/Merge Mobility Number of data packets TCP Retransmit Timeouts

In 54 trials. . . 9.25% false-positives (type I error). 1.85% false-negatives (type II error).

Kyle Usbeck Network-Aware Automated Planning 71/75

slide-155
SLIDE 155

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Outline

1 Introduction Motivation Background Approach 2 Formalization Problem Statement 3 Technical Approach Planning Agents Execution Agents Monitoring Agents Mixed-initiative UI 4 Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Kyle Usbeck Network-Aware Automated Planning 72/75

slide-156
SLIDE 156

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Main Contributions

1

Qualitatively-different plan generation:

Qualitatively different plans over a range of plan evaluation criteria. Visualizing plan evaluations.

2

Network-aware agents:

Network-aware planning agent. Network-aware execution agents. Network-aware monitoring agents.

Kyle Usbeck Network-Aware Automated Planning 73/75

slide-157
SLIDE 157

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Main Contributions

1

Qualitatively-different plan generation:

Qualitatively different plans over a range of plan evaluation criteria. Visualizing plan evaluations.

2

Network-aware agents:

Network-aware planning agent. Network-aware execution agents. Network-aware monitoring agents.

Kyle Usbeck Network-Aware Automated Planning 73/75

slide-158
SLIDE 158

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Main Contributions

1

Qualitatively-different plan generation:

Qualitatively different plans over a range of plan evaluation criteria. Visualizing plan evaluations.

2

Network-aware agents:

Network-aware planning agent. Network-aware execution agents. Network-aware monitoring agents.

Kyle Usbeck Network-Aware Automated Planning 73/75

slide-159
SLIDE 159

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Main Contributions

1

Qualitatively-different plan generation:

Qualitatively different plans over a range of plan evaluation criteria. Visualizing plan evaluations.

2

Network-aware agents:

Network-aware planning agent. Network-aware execution agents. Network-aware monitoring agents.

Kyle Usbeck Network-Aware Automated Planning 73/75

slide-160
SLIDE 160

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Main Contributions

1

Qualitatively-different plan generation:

Qualitatively different plans over a range of plan evaluation criteria. Visualizing plan evaluations.

2

Network-aware agents:

Network-aware planning agent. Network-aware execution agents. Network-aware monitoring agents.

Kyle Usbeck Network-Aware Automated Planning 73/75

slide-161
SLIDE 161

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Main Contributions

1

Qualitatively-different plan generation:

Qualitatively different plans over a range of plan evaluation criteria. Visualizing plan evaluations.

2

Network-aware agents:

Network-aware planning agent. Network-aware execution agents. Network-aware monitoring agents.

Kyle Usbeck Network-Aware Automated Planning 73/75

slide-162
SLIDE 162

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Main Contributions

1

Qualitatively-different plan generation:

Qualitatively different plans over a range of plan evaluation criteria. Visualizing plan evaluations.

2

Network-aware agents:

Network-aware planning agent. Network-aware execution agents. Network-aware monitoring agents.

Kyle Usbeck Network-Aware Automated Planning 73/75

slide-163
SLIDE 163

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Future Work

Knowledge-based monitoring agents. Incorporate the effects of planning actions into heuristics.

Kyle Usbeck Network-Aware Automated Planning 74/75

slide-164
SLIDE 164

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Future Work

Knowledge-based monitoring agents. Incorporate the effects of planning actions into heuristics.

Kyle Usbeck Network-Aware Automated Planning 74/75

slide-165
SLIDE 165

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Acknowledgements

Advisor:

  • Dr. William C. Regli

Committee:

  • Dr. Rachel Greenstadt
  • Dr. Ani Hsieh

Conceptual Contributors:

  • Prof. Austin Tate
  • Dr. Gerhard Wickler

Jeff Dalton

Co-workers Critiquors:

Ilya Braude Matt Chase Patrick Freestone Joe Kopena Duc Nguyen Rob Lass Evan Sultanik

Family & Friends L

AT

EX, Vim, opensource software

Kyle Usbeck Network-Aware Automated Planning 75/75

slide-166
SLIDE 166

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

IED Detection Accuracy and Bandwidth Usage

2 4 6 8 10 0.2 0.4 0.6 0.8 1 Network Bandwidth Usage (Mbps) IED Detection Accuracy (%) IED Detection Accuracy vs. Network Bandwidth Usage Guided I-Plan Default Random

Kyle Usbeck Network-Aware Automated Planning 76/75

slide-167
SLIDE 167

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

IED Detection Accuracy and Execution Time

10000 20000 30000 40000 50000 60000 70000 0.2 0.4 0.6 0.8 1 Plan Execution Time (sec) IED Detection Accuracy (%) IED Detection Accuracy vs. Plan Execution Time Guided I-Plan Default Random

Kyle Usbeck Network-Aware Automated Planning 77/75

slide-168
SLIDE 168

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Network Bandwidth Usage and Execution Time

10000 20000 30000 40000 50000 60000 70000 2 4 6 8 10 Plan Execution Time (sec) Network Bandwidth Usage (Mbps) Network Bandwidth Usage vs. Plan Execution Time Guided I-Plan Default Random

Kyle Usbeck Network-Aware Automated Planning 78/75

slide-169
SLIDE 169

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Network Hops and IED Detection Accuracy

0.2 0.4 0.6 0.8 1 2 4 6 8 10 IED Detection Accuracy (%) Number of Network Hops (full data path) Network Hops vs. IED Detection Accuracy Guided I-Plan Default Random

Kyle Usbeck Network-Aware Automated Planning 79/75

slide-170
SLIDE 170

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Network Hops and Bandwidth Usage

2 4 6 8 10 2 4 6 8 10 Network Bandwidth Usage (Mbps) Number of Network Hops (full data path) Network Hops vs. Network Bandwidth Usage Guided I-Plan Default Random

Kyle Usbeck Network-Aware Automated Planning 80/75

slide-171
SLIDE 171

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Network Hops and Execution Time

10000 20000 30000 40000 50000 60000 70000 2 4 6 8 10 Plan Execution Time (sec) Number of Network Hops (full data path) Network Hops vs. Plan Execution Time Guided I-Plan Default Random

Kyle Usbeck Network-Aware Automated Planning 81/75

slide-172
SLIDE 172

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Eval. Benchmarking Execution Time Distribution

5 10 15 20 25 30 21381 25519 29657 33796 37934 42073 46211 50350 54488 58627 62765 guided random iplan

Kyle Usbeck Network-Aware Automated Planning 82/75

slide-173
SLIDE 173

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Eval. Benchmarking IED Detect. Acc. Distribution

5 10 15 20 25 1 2 3 4 5 6 7 8 9 10 guided random iplan

Kyle Usbeck Network-Aware Automated Planning 83/75

slide-174
SLIDE 174

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Eval. Benchmarking Link Quality Distribution

5 10 15 20 25 30 1 2 3 4 5 6 7 8 9 guided random iplan

Kyle Usbeck Network-Aware Automated Planning 84/75

slide-175
SLIDE 175

Computer

Department of Science

Introduction Formalization Technical Approach Experiments Plan Evaluation Benchmarking Network-Aware Agent Combinations Discussion

Plan Eval. Benchmarking Bandwidth Usage Distribution

5 10 15 20 25 30 3.1 3.7 4.3 4.8 5.4 6.0 6.6 7.1 7.7 8.3 guided random iplan

Kyle Usbeck Network-Aware Automated Planning 85/75