Http:// European Network of Excellence in AI Planning PLANET Gap Bridging Seminar (GBS-2) Prof. dr. Tim Grant Atos Origin Nederland

http://www.planet-noe.org

European Networkof Excellence inAI Planning

PLANETGap Bridging Seminar (GBS-2)

Prof. dr. Tim Grant

Atos Origin Nederland / University of [email protected] / [email protected]

2PLANET GBS-2, Delft, 21 Nov 02

Overview

• Presentation goal:– To introduce 2nd PLANET Gap-Bridging Seminar

• Structure:– Purpose of GBS-2– Our sponsors– EU Network of Excellence in Planning (PLANET)– Features & benefits analysis– Introducing the speakers


GBS-2: purpose (1)

• Motivation for GBS-2:– Academia & industry both work in P&S– Their goals differ:

• Academia:– Develops new knowledge & technologies

– Performance measured in terms of publications

• Industry:– Applies knowledge & technologies in products & services

– Performance measured in financial terms

– Communications gap!


GBS-2: purpose (2)

• GBS-2’s aim:– To bridge the gap between academia & industry

• Approach:– Industrial speakers talk about real-world applications– Academic audience

• Benefits:– To audience:

• Gain understanding of wider influences on P&S• Gain new problems & insights that can guide R&D

– To speakers:• Visibility, state-of-the-art, partnerships, recruitment


GBS-2: sponsors

• PLANET:– EU Network of Excellence in AI Planning– http://www.planet-noe.org/

• SIKS:– Dutch research school for Information & Knowledge Systems– http://www.siks.nl/

• TU Delft:– Leading Dutch university of technology– http://www.tudelft.nl

• Atos Origin:– Leading international IT services provider– http://www.atosorigin.com/


PLANET (1)

• Overall aim:– Build co-ordinated network to support research,

development, & technology transfer in Europe

• Objectives:– Encourage pan-European co-operation between

research groups & industry– Promote technology transfer by early

involvement of industrial partners– Co-ordinate training & exchange of researchers

and people from industry & public administration– Provide an information infrastructure and a

representative source of expertise


PLANET (2)

Research and Development Co-ordinationMalik Ghallab, Lee McCluskey

1

Training and ExchangeSam Steel, Susanne Biundo

4

Technology Transfer and CommunicationRuth Aylett, Tim Grant, Brian Drabble

2 Technical Co-ordination Units (TCUs)Amedeo Cesta, Alfredo Milani

3

Information Infrastructureand ManagementSusanne Biundo, Bernd Schattenberg

5

• Work packages:


PLANET (3)

• Activities:– General network meetings– TCU meetings/workshops (often linked to conferences)– Cross-site visits (between PLANET members)– Information days (for industry)– Summer schools (for students)– Gap-bridging seminars

• Deliverables:– Software catalogue & repository– Benchmarks & case studies– Newsletters, posters, TCU flyers– Annual reports– P&S curriculum– Exploitation plan

All downloadable fromhttp://www.planet-noe.org/


PLANET (4)

• Technical Co-ordination Units (TCUs):– Identify areas of current needs & opportunities amongst

PLANET members

AerospaceApplicationsAmedeo CestaTim Grant

WorkflowManagementDaniel Borrajo

Robot PlanningMalik Ghallab

Knowledge EngineeringLee McCluskey

Intelligent ManufacturingRuth Aylett

On-line Planningand SchedulingGerard Verfaillie

Web-based ApplicationsAlfredo Milani


PLANET (5)

• TCU road-maps identify:– Current state of P&S technology– Recent developments– Relationships to other fields– Application problems & deficiencies– Technology transfer issues– Directions for potential exploitation– Application-driven improvements– Plans for focussed actions

Downloadable fromhttp://www.planet-noe.org/


Features & benefits (1)

• PLANET ends August 2003• Now starting Exploitation Plan:

– Short term (3 years)– Medium term (6 years)– Long term (10 years)

• Marketing analysis proposed to aid exploitation:– Stakeholder analysis– Survey of European capabilities & offerings– SWOT analysis– Features & benefits analysis– Marketing plan & marketing materials


Features & benefits (2)Requirements

need

Design

specification

Manufacture

design

Operations

new system

Decommission

old systemOps prep

previousexperience

SOPs,trained ops

Planning Controlplans

raw materialsproducts/services

current state & past experience

OBSERVE

ORIENT

DECIDE

ACT



• Features & benefits analysis:– Aim:

• To define product / service in customer’s terms

– Feature:• What product / service has or is

– Benefit:• What product / service does for customer

– Analysis procedure:• Identify customer populations & their needs• Decide what product / service is to be sold• Identify features that product / service provides• Match features to benefits that customers obtain• Prove that benefits can be delivered



• AI P&S products & services:– Some possible products:

• Planning(-domain) knowledge acquisition software• Plan-generation software• Plan-animation software• Plan-evaluation software• Plan-execution software• Plan-recognition software

– Some possible services:• Planning software selection, installation & integration• Tailoring planning software specific to domain• Domain knowledge capture & representation• Plan execution, monitoring & control



• Plan-generator can be:– Human:

• Unaided• Supported by non-intelligent software:

– Spreadsheet, database, planning package

• Supported by intelligent software (“mixed-initiative planning”)

– Automation:• Hardware (invariably non-intelligent, e.g. PLCs)• Non-intelligent software (“canned” responses)• Intelligent software

• Comparisons:– Intelligent plan-generation software vs. human– Intelligent plan-generation software vs. hardware– Intelligent vs. non-intelligent plan-generation software

Existing methods

What AI planning community is working on



• Example features of plan-generator:– Functional:

• It uses domain knowledge (including domain constraints)• It can decompose goals• It puts actions into sequences that meet domain constraints• It generates sequences from initial to goal state• It selects suitable (renewable) resources for actions• It can reason at multiple levels of (domain) abstraction• It can reason about concurrent action• It can handle uncertain information• Given same problem, it is guaranteed to produce same plan

– Non-functional:• It is tireless (24 hours/day, 365 days/year)• It is reliable (e.g. 10-5 availability over one year)• It works in real-time; it can generate valid plan by a deadline• It integrates to customer’s other IT systems• It provides standard (system, user) interfaces



• Possible customer benefits:– Faster:

• Installed and/or tailored more quickly• Generates plan more quickly• Updates plan more quickly after change

– Better:• Plans for more complex domains• Generates better-quality plans (less domain-resource usage)• Generates plans that have fewer errors• Generates plans that are more robust to changes

– Cheaper:• Lower purchase cost• Lower running costs:

– Fewer operators needed– Less operator training needed– Less hardware / cheaper hardware



• Real-world proof:– Brian Drabble, University of Oregon/ OnTime Inc.:

• Submarine construction (Electric Boat):– 1.4 million tasks– 30% reduction in overtime & subcontracting– Runtime 1 day instead of 6 weeks

• Aircraft construction (Boeing):– 570 tasks, 17 resources– 10-15% shorter, saving US$1.2 million per aircraft– Runtime 30 mins (2 orders of magnitude faster)

– ARPA Planning Initiative:• Logistics build-up to Gulf War, 1991:

– Saved more than the cost of all US AI research– Sources: Cross & Walker (1994) & AIPS’96


GBS-2: speakers

• Alessandro Donati:– “Space Mission Operations P&S: Past, present, & future”– European Space Operations Centre– http://www.esoc.esa.de

• Henk Hesselink:– “AI Planning: waiting for results?”– Dutch National Aerospace Laboratory (NLR)– http://www.nlr.nl

• Frank Oxener:– “Work Force Planning: A logical next step after ERP”– Atos Origin Nederland b.v.– http://www.atosorigin.com

• Yossi Rissin & Roman Bartak:– “When Theory Crashed into Reality”– Visopt b.v.– http://www.visopt.com

Http:// European Network of Excellence in AI Planning PLANET Gap Bridging Seminar (GBS-2) Prof. dr. Tim Grant Atos Origin Nederland

Documents

planet membersplanet

purpose of gbs

recruitmentplanet gbs

speakersplanet gbs

planet gap

planet membersinformation

coordinated network

product service