All rights reserved, California Institute of Technology © 2002 Argumentation for Coordinating Shared Activities (a talk on distributed planning) Brad Clement,

All rights reserved, California Institute of Technology © 2002

Argumentation for Coordinating Shared Activities

(a talk on distributed planning) Brad Clement, Tony Barrett, Steve Schaffer

Artificial Intelligence Group

Jet Propulsion Laboratory

California Institute of Technology{bclement,barrett,srschaff}@aig.jpl.nasa.gov

http://www-aig.jpl.nasa.gov/


MotivationOver 40 multi-spacecraft missions proposed!

– Autonomous single spacecraft missions have not yet reached maturity.

– How can we cost-effectively manage multiple spacecraft?

Earth Observing System Sun-Earth Connections

Origins Program

Structure & Evolution of the Universe

Mars Network

NMP

NMP


Motivation

• Considerable ground operations effort and cost involved in coordinating mission plans for interacting missions.

• Human collaboration can be error-prone and slow to react.

• Automating this coordination reduces operations costs and increases science return.

• On board a team of spacecraft, it can be too expensive to centralize information and processing


Why Decentralized Planning?

• Why plan?– near-term actions can effect subsequent ones in

achieving longer-term goals

• Why decentralize?– competing objectives (self-interest)– control is already distributed– communication constraints/costs (b/w, delay, privacy)– computation constraints (parallel processing)– robustness to failure?


Prior Work

• Treats decentralized planning as an offline, collaborative problem– planners collaborate on resolving state conflicts,

ignore communication costs

• Space missions present real-time problems with self-interested agents– scientists compete for instrument/spacecraft use– missions compete for bandwidth to Earth– remote explorers may need to respond to

dynamics autonomously


Problems

• How should planning agents communicate with each other?– shared activities

• How can they coordinate joint actions during execution?– continual coordination algorithm– consensus window

• How can coordination algorithms be developed efficiently?– protocol classes that manipulate shared activities


Shared Activity Coordination (SHAC)

– continual coordination algorithm– language for coordinating planning agents– framework for defining and implementing automated

interactions between planning agents (a.k.a. coordination protocols/algorithms)

– software• planner-independent interface• protocol class hierarchy• testbed for evaluating protocols


ExecutiveExecutive

Planner

ExecutiveExecutive

Planner

ExecutiveExecutive

Planner

Shared Activity Coordination

Shared activities implement team plans, joint actions, and shared states/resources


SHAC Applications• Simulated Mars network

– Detailed s/c model– Coordination in real time– Restricted communication (orbital

constraints)– Focus on communication quality of

service

• MISUS – mechanism for delegating goals for a team of rovers

• Techsat-21– Coordinating ground planning– Abandoned when mission de-

scoped

• Deep Space Network resource allocation (future)

MGS MEX Odyssey

MER A MER B

Mission Planning

Simulation Env

Commanding SOH displayTelemetry

ASPEN

SCL

Fight Dynamics

Payload Ops W/S

Cmd Verification Engineering Models

PPC ClusterCmd Verification

TT&C W/S TT&C W/S

Data Center

Pass PlaybackSOH displayTrendingAnom Res

SCLMatlab

http://nssdc.gsfc.nasa.gov/thumbnail/spacecraft/mars_express.gif


Shared Activity Model

• parameters (string, integer, etc.)– constraints (e.g. agent4 allows start_time [0,20], [40,50])

• decompositions (shared subplans)

• permissions - to modify parameters, move, add, delete, choose decomposition, constrain

• roles - maps each agent to a local activity

• protocols - defined for each role– change constraints– change permissions– change roles

• includes adding/removing agents assigned to activity


Argumentation

• Proposals and counterproposals with justifications

• In distributed constraint satisfaction– Proposals are variable changes– Justifications are no-goods

• For distributed planning– Proposals are shared activity changes– Justifications are constraints


SHAC AlgorithmGiven: a plan with multiple activities, including a set of

shared_activities, and a projection of plan into the future.1. Revise projection using the currently perceived state and any

newly added goal activities.2. Alter plan and projection while honoring constraints and

permissions of shared_activities.3. Release relevant near-term activities of plan to the real-time

execution system.4. For each shared activity in shared_activities

– apply each associated protocol to modify the activity5. Communicate changes in shared_activities.6. Update shared_activities based on received communications.7. Go to 1.


Protocol CapabilitiesDefining/extending protocol classes1. modify permissions2. modify local parameter constraints3. add/delete sharing agents4. change roles of sharing agents

Default protocol class• joint intention• mutual belief• resource sharing• active/passive roles• master/slave roles


Control Protocols for a Shared Activity

• Chaos– A free-for-all among planners

• Master/Slave– The master has permissions, slaves don’t

• Round Robin– Master role passes round-robin among planners

• Asynchronous Weak Commitment (AWC)– Neediest planner becomes master

• Variations– how many planners share activity

– use of constraints


Experiments – Abstract Problem

• joint measurements

• capability matching

• 3-9 spacecraft

• 1-7 capabilities

• 1-9 joint goals each requiring 1-4 of each capability


Chaos - invalid solutions

M/S - not complete

Experimental Results(Progress over cpu time)

num

ber

of p

robl

ems

max cpu time (seconds)

AWC

RR

ChaosM/S


num

ber

of p

robl

ems

max actual time (seconds)

AWC

RR

ChaosM/S

Experimental Results(Progress over clock time)

M/S – solves quickly or not at all


num

ber

of p

robl

ems

max number of messages

AWC

RR

ChaosM/S

Experimental Results(Number of messages sent for problems solved)

M/S – order(s) of magnitude fewer messages sent

RR – performance flip flops with rest


num

ber

of p

robl

ems

max data sent (bytes)

AWC

RR

Chaos

M/S

Experimental Results(Data sent for problems solved)

M/S – order(s) of magnitude less data sent

Performance flip flops for rest


Experimental Results - Sharing(Progress over cpu time)

num

ber

of p

robl

ems

max cpu time (seconds)

AWCRR

Chaos-BM/S-B

AWC-B

M/SChaosRR-B

Share with all (broadcast) or share only with assigned s/c

RR – performance best without and worst with broadcast

Chaos – much better with broadcast


Experimental Results - Sharing(Progress over clock time)

num

ber

of p

robl

ems

max actual time (seconds)

AWC

RR

Chaos-B

M/S-B

AWC-B

M/S

Chaos

RR-B


RR – performance best without and worst with broadcast

Chaos – much better with broadcast


Experimental Results - Sharing(Number of messages sent for problems solved)

num

ber

of p

robl

ems

max number of messages

AWC

RR

Chaos-B

M/S-B

AWC-B

M/SChaos

RR-B


Many order of magnitude separations

In general, protocols that solve more problems send more messages

RR – performs much better on “hardest” 500 problems


Experimental Results - Sharing(Data sent for problems solved)

num

ber

of p

robl

ems

max data sent (bytes)

AWC

RR

Chaos-B

M/S-B

AWC-B

M/SChaos

RR-B


Many order of magnitude separations

In general, protocols that solve more problems send more data

RR – performs much better on “hardest” 500 problems


Summary• SHAC

– communication language for distributed planning

– general algorithm for continual coordination

– framework for developing coordination protocols

– software with planner independent interface

• Characteristics and performance of argumentation-based protocols– Round-robin (with limited sharing) performed fastest with

somewhat heavy communication costs

– AWC is all around best with high communication costs

– M/S has least communication costs but only works for restricted domains


Future Directions

• evaluate other simple protocols in other domains

• different constraint representations• abstraction techniques for limiting

communication and preserving flexibility• use group communication techniques to give

consistency guarantees to protocols like chaos

• find a customer

All rights reserved, California Institute of Technology © 2002 Argumentation for Coordinating Shared Activities (a talk on distributed planning) Brad Clement,

Documents

processingall rights

shared activitiesall

multispacecraft missions

coordination algorithms

shared activitieshow

team of spacecraft

multiple spacecraft

plannersmany plans