Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, CAMERA Coordination and Management...

18
Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/ CAMERA CAMERA Coordination and Management Coordination and Management Environments for Responsive Agents Environments for Responsive Agents Robert Neches, P/I Robert Neches, P/I Pedro Szekely, Project Pedro Szekely, Project Leader Leader Martin Frank, Comp. Martin Frank, Comp. Scientist Scientist University of Southern California University of Southern California Information Sciences Institute Information Sciences Institute Marina del Rey, California Marina del Rey, California Capt Dowell Capt Dowell SSgt Fray SSgt Fray Capt Greene Capt Greene Capt Hardy Capt Hardy Capt Maddocks Capt Maddocks Maj Oneill Maj Oneill Sgt Thompkins Sgt Thompkins Maj Vest Maj Vest MAG 13, 3RD MAW MAG 13, 3RD MAW

Transcript of Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, CAMERA Coordination and Management...

Page 1: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

CAMERACAMERA

Coordination and Management Coordination and Management Environments for Responsive AgentsEnvironments for Responsive Agents

Robert Neches, P/IRobert Neches, P/IPedro Szekely, Project LeaderPedro Szekely, Project Leader

Martin Frank, Comp. ScientistMartin Frank, Comp. Scientist

University of Southern CaliforniaUniversity of Southern CaliforniaInformation Sciences InstituteInformation Sciences Institute

Marina del Rey, CaliforniaMarina del Rey, California

Robert Neches, P/IRobert Neches, P/IPedro Szekely, Project LeaderPedro Szekely, Project Leader

Martin Frank, Comp. ScientistMartin Frank, Comp. Scientist

University of Southern CaliforniaUniversity of Southern CaliforniaInformation Sciences InstituteInformation Sciences Institute

Marina del Rey, CaliforniaMarina del Rey, California

Capt DowellCapt DowellSSgt FraySSgt Fray

Capt GreeneCapt GreeneCapt HardyCapt Hardy

Capt MaddocksCapt MaddocksMaj OneillMaj Oneill

Sgt ThompkinsSgt ThompkinsMaj VestMaj Vest

MAG 13, 3RD MAWMAG 13, 3RD MAW

Capt DowellCapt DowellSSgt FraySSgt Fray

Capt GreeneCapt GreeneCapt HardyCapt Hardy

Capt MaddocksCapt MaddocksMaj OneillMaj Oneill

Sgt ThompkinsSgt ThompkinsMaj VestMaj Vest

MAG 13, 3RD MAWMAG 13, 3RD MAW

Page 2: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

SNAP: Schedules Negotiated SNAP: Schedules Negotiated by ANT-based Plannersby ANT-based Planners

FlightFlightSchedulingScheduling

SNAPSNAP Develops weekly and daily schedulesDevelops weekly and daily schedules

Repairs schedules on-the-fly for unforeseen eventsRepairs schedules on-the-fly for unforeseen events

Allows operators to refine the scheduleAllows operators to refine the schedule

Interacts with SARA for risk analysis, record-keepingInteracts with SARA for risk analysis, record-keeping

SNAP ANTsSNAP ANTs PilotsPilots

AircraftAircraft

MissionsMissions

RangesRanges

Flow managerFlow manager

Add’l futureAdd’l future ODOODO OrdnanceOrdnance PMCPPMCP AcademicsAcademics

ANTsANTs Represent the objectives and Represent the objectives and

concerns of stakeholdersconcerns of stakeholders

NegotiationNegotiation Enables rapid identification of Enables rapid identification of

candidates to perform taskscandidates to perform tasks Aids investigating trade-offsAids investigating trade-offs

Contribution: coordination of skills,Contribution: coordination of skills,training, missions, equipment, and timetraining, missions, equipment, and time

Page 3: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

GuidanceGuidance

Resource AllocationResource AllocationNegotiationsNegotiations

Progress MonitoringProgress Monitoring

CandidateCandidateTasksTasks

TaskTaskExecutionExecution

Ro

ute

d G

uid

an

ceR

ou

ted

Gu

ida

nce

Negotiation StatusNegotiation Status

Elab

oratio

ns to

Gu

ida

nce

Elab

oratio

ns to

Gu

ida

nce

Pu

shb

acks o

n G

uid

an

ceP

ush

back

s on

Gu

ida

nce

StatusStatus

CAMERA: CAMERA: Approach to Real-time, Approach to Real-time, Distributed Resource AllocationDistributed Resource Allocation

Application: Coordinated Ops, Maintenance, SupplyApplication: Coordinated Ops, Maintenance, Supply

Different from traditional optimizationDifferent from traditional optimization

– Cannot elicit full evaluation function in advanceCannot elicit full evaluation function in advance

– Often, only way to handle problem is to change itOften, only way to handle problem is to change it

VERY high military payoffs: Money, LivesVERY high military payoffs: Money, Lives

Negotiation costs increase near execution time (aids closure)Negotiation costs increase near execution time (aids closure)

Time permitting, refine priorities by exploring options outside of guidanceTime permitting, refine priorities by exploring options outside of guidance

Drive convergence by adding guidance to steer negotiations within original guidanceDrive convergence by adding guidance to steer negotiations within original guidance

Guide convergence by soft bounds on counterproposals: time, cost, variation

Guide convergence by soft bounds on counterproposals: time, cost, variation

Multiple feedback loops control the negotiation process

Multiple feedback loops control the negotiation process

Parallel negotiation of alternatives, pruned by guidance sensitive to time pressures (aids efficiency, stability)Parallel negotiation of alternatives, pruned by guidance sensitive to time pressures (aids efficiency, stability)

Execute current best alternatives when time arrives (ensures good-enough, soon-enough)

Execute current best alternatives when time arrives (ensures good-enough, soon-enough)

Deviations from expected progress trigger renegotiation in anticipation of problems, opportunities (aids stability)Deviations from expected progress trigger renegotiation in anticipation of problems, opportunities (aids stability)

Page 4: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

GuidanceGuidance

Resource AllocationResource AllocationNegotiationsNegotiations

Progress MonitoringProgress Monitoring

CandidateCandidateTasksTasks

TaskTaskExecutionExecution

Ro

ute

d G

uid

an

ceR

ou

ted

Gu

ida

nce

Negotiation StatusNegotiation Status

Elab

oratio

ns to

Gu

ida

nce

Elab

oratio

ns to

Gu

ida

nce

Pu

shb

acks o

n G

uid

an

ceP

ush

back

s on

Gu

ida

nce

StatusStatus

CAMERA Status:CAMERA Status:Currently Implemented Currently Implemented

Architectural ComponentsArchitectural Components

~300 ANTs~300 ANTs

~50K msgs ~50K msgs in < 30 secsin < 30 secs

~250 tasks ~250 tasks consideredconsidered

~ 50 tasks ~ 50 tasks acceptedaccepted

Page 5: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

JSF Comments on Key Technology:JSF Comments on Key Technology:Major General Hough, JSF Program ManagerMajor General Hough, JSF Program Manager

“... a first time opportunity to do extremely complex

balancing of considerations within critical time constraints,

as well as repair those schedules within those same

time limits.”

“What excited me most, though, was the empowerment...to do

planning and replanning so far inside decision cycles that it will

provide “Look-ahead” capabilities never before

available.”

Page 6: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Themes in Current Phase of WorkThemes in Current Phase of Work

Architectural extensions and refinements

Initial applicability at MCAS Yuma

Collaboration with complexity contractors

Hierarchical negotiation: SNAP / MAPLANT

Page 7: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

CAMERA:CAMERA:Architecture Work in ProgressArchitecture Work in Progress

Dynamically Dynamically introduced introduced constraintsconstraints

Soft boundsSoft bounds

““Marbles”Marbles”

Parallel Parallel evaluationevaluation

Architectural work remaining in phase III: proactive renegotiationArchitectural work remaining in phase III: proactive renegotiation

GuidanceGuidance

Resource AllocationResource AllocationNegotiationsNegotiations

Progress MonitoringProgress Monitoring

CandidateCandidateTasksTasks

TaskTaskExecutionExecution

Ro

ute

d G

uid

an

ceR

ou

ted

Gu

ida

nce

Negotiation StatusNegotiation Status

Elab

oratio

ns to

Gu

ida

nce

Elab

oratio

ns to

Gu

ida

nce

Pu

shb

acks o

n G

uid

an

ceP

ush

back

s on

Gu

ida

nce

StatusStatus

Initiate exploration of tradeoffs and alternatives by imposing and freezing requirements and their priorities

Initiate exploration of tradeoffs and alternatives by imposing and freezing requirements and their priorities

Control exploration via measurable limits on deviation from idealControl exploration via measurable limits on deviation from ideal

Trigger cooperative renegotiation over critical resources through continuous evaluation of needTrigger cooperative renegotiation over critical resources through continuous evaluation of need

Evaluate tradeoffs of selected alternatives through negotiation “beam search”Evaluate tradeoffs of selected alternatives through negotiation “beam search”

Page 8: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Page 9: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Page 10: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Page 11: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Priorities regroupedPriorities regrouped

Mission selected for editingMission selected for editing

Page 12: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Page 13: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Feedback from Users at Marine Air Group 13:Feedback from Users at Marine Air Group 13:Commander of VMA-513, Designated Representative for Evaluating CACE ACTDCommander of VMA-513, Designated Representative for Evaluating CACE ACTD

“... this tool is not simply an automation of current processes..., but a catalyst for the

expansion of a squadron’s planning horizon that affords the comparison of alternative ‘futures.’”

“...I am encouraged by the progress our that our military- civilian team team has made...”

When perfected, it should allow commanders to apply, evaluate..., and refine command guidance; bringing an unparalleled level of detail and

utility to operational decisions.

...Equally useful is the ability to repair plans that have been overcome by events... without

deviating from the commander’s intent.

Page 14: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Marbles: Marbles: Cooperative RenegotiationCooperative Renegotiation

Dynamically Dynamically introduced introduced constraintsconstraints

Soft boundsSoft bounds

““Marbles”Marbles”

Parallel Parallel evaluationevaluation

GuidanceGuidance

Resource AllocationResource AllocationNegotiationsNegotiations

Progress MonitoringProgress Monitoring

CandidateCandidateTasksTasks

TaskTaskExecutionExecution

Ro

ute

d G

uid

an

ceR

ou

ted

Gu

ida

nce

Negotiation StatusNegotiation StatusE

labo

ration

s to G

uid

an

ceE

labo

ration

s to G

uid

an

ce

Pu

shb

acks o

n G

uid

an

ceP

ush

back

s on

Gu

ida

nce

StatusStatus

Trigger cooperative renegotiation over critical resources through continuous evaluation of needTrigger cooperative renegotiation over critical resources through continuous evaluation of need

Page 15: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Marbles Family of Resource Allocation NegotiationsMarbles Family of Resource Allocation Negotiations

Task priorities = discrete quantities to put on resources Quantities adapt as execution time nears Continuous fluctuation Graceful withdrawal

Task priorities = discrete quantities to put on resources Quantities adapt as execution time nears Continuous fluctuation Graceful withdrawal

Mission 2Lead Pilot w/ Quals QL21, QL22....

Wing w/ Quals QW21, QW22....

Pilot APilot B

Pilot C

Pilot D

Mission 1Lead Pilot w/ Quals QL11, QL12....

Wing w/ Quals QW11, QW12....

Page 16: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Study of Marbles Family MembersStudy of Marbles Family Members Example: comparisons on 30

randomly generated problems of total value (quality), time (speed)

3 Marbles variants vs. Simulated Annealing (SA) and SAT– MarbleSize: allows subdividing

marbles, trades solution quality vs convergence speed

– GrabMarbles: like MarbleSize but allows local heuristics

– Marbles2: dutch auction-like (loser)

Results / Observations:Results / Observations:– MarbleSize & GrabMarbles work MarbleSize & GrabMarbles work

as well as SA and SATas well as SA and SAT

– Run 3-10x fasterRun 3-10x faster

– Appear tunable toAppear tunable toCAMERA and ANTCAMERA and ANTgood-enough / soon-enough good-enough / soon-enough criteriacriteria

Page 17: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

Computational Complexity IssuesComputational Complexity Issues Engineering handbook theory of negotiation:

– No best algorithm

– Real question is, “Best under which conditions?”

– Domain knowledge is critical

Next step in CAMERA: parallel evaluation– ATTEND has results that could significantly improve this

Remnants of “Complexity Fest” workshop series– CAMERA code available to complexity contractors

– Hooks implemented for monitoring negotiations

– Other contractors’ plans to be presented during this mtg

Page 18: Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA,  CAMERA Coordination and Management Environments for Responsive Agents.

Robert Neches & Pedro Szekely, USC ISI, © 2001 CAMERA, http://www.isi.edu/camera/

SNAP - MAPLANT CoordinationSNAP - MAPLANT Coordination

OPSMaintenance

Guidance

MAPLANT SNAP

First Cut Plan

Refined Ops Plan

Approx. Maintenance Plan (A/C status)

Refined Maintenance Plan

DemoDemo

Negotiation