1 hedlamp kick-off meeting schlumberger cambridge, 3/4/2012 hedlamp (huddersfield + edinburgh -...

1

HEDLAMP Kick-Off Meeting

Schlumberger Cambridge, 3/4/2012

HEDLAMP (Huddersfield + HEDLAMP (Huddersfield + Edinburgh - Learning and Edinburgh - Learning and Adaptation of Models for Adaptation of Models for

Planning)Planning)

Gerhard WicklerGerhard WicklerAIAI, University of Edinburgh, UKAIAI, University of Edinburgh, UK

[email protected]@ed.ac.uk

2



Introduction / OverviewIntroduction / Overview

AI Planning at Edinburgh:– Early Work on HTN Planning (70s): Austin Tate’s NONLIN

– The Open Planning Architecture: O-Plan (80s & 90s)

– New Technology: I-X (2000s)

AIAI and I-X:– I-X used in many AIAI planning-related projects (open

source)

– I-X applications: domain-specific; possibly restricted distribution

Overview (this presentation):– Generic HTN Planning Technology: I-X and the Process

Panel

– Applications: I-X and Its Use in Emergency Response

– Procedural Knowledge Engineering with Domain Features

– I-Room: Collaboration Support in Virtual Worlds

3



HEDLAMP PartnersHEDLAMP Partners

University of Huddersfield - Department of Informatics - Knowledge Engineering and Intelligent Interfaces research group: Prof. Lee McCluskey and Dr. Peter Gregory

University of Edinburgh - School of Informatics - CISA/AIAI: Prof. Austin Tate and Dr. Gerhard Wickler

Huddersfield– Domain Model Generation– Machine Learning

Edinburgh:– Rich Plan Generation– Dynamic Multi-Agent Planning– Mixed-Initiative Operations Centres

http://www.aiai.ed.ac.uk/project/hedlamp/

4



““To-Do” Lists and To-Do” Lists and I-X Process PanelsI-X Process Panels

organizing work: to-do lists– universally acknowledged to be useful– help people remembering what needs to be

done» ensure everything gets done» provide overview of unaccomplished tasks

the I-X framework– for creating applications in which multiple

agents » adopt a task-centric view of a situation» must coordinate to perform their activities

I-X Process Panels– principal interface to an I-X application– provide functionality of to-do list (amongst

others)

5



I-X Process Panel: I-X Process Panel: ExampleExample

issues: questions activities: items on the

to-do list– described as verb plus

parameters

– can be annotated

– have priority

– actions suggested by I-X

» tick off

» refine task

» pass to others

» use custom handler

state: world state information

annotations: none shown

6



The I-X Tool Suite for e-The I-X Tool Suite for e-ResponseResponse

I-Space agent

relationship/ capability

tool

I-Space agent

relationship/ capability

tool

Presence Tool agent

presence awareness

Presence Tool agent

presence awareness

Map Tool Situation

awareness

Map Tool Situation

awarenessWeb

Resources SOP

repository

Web Resources

SOP repository

3-D Viewer3-D Viewer

Domain Editor

create and manage

SOPs

Domain Editor

create and manage

SOPs

PDA Viewer

PDA Viewer

Message Tool

formal/informal comms

channel

Message Tool

formal/informal comms

channel

Process Panel

Principal activity/ issue

interface

Process Panel

Principal activity/ issue

interface

7



Co-OPR2Co-OPR2

Collaboration and Communication

CommandCentre

IsolatedPersonnel

EmergencyResponders

CentralAuthorities

8



FireGridFireGrid

Data access: maps,

models

Super-real-time simulation (HPC)

Command-and-Control

Emergency

responders

1000s of sensors & gateway processing

GridGrid

I-X Technologie

s

9



OpenVCEOpenVCE

OpenVCE: supports collaboration in virtual spaces– website based on open source content

management system (Drupal + MediaWiki) for asynchronous collaboration

– 3D virtual space for synchronous collaboration (Second Life)

– virtual collaboration protocol: SOP for using OpenVCE

10



Further BackgroundFurther Background

Rich Plan Representation– <I-N-C-A> Plan Abstraction– MIT Process Handbook – PIF– NIST PSL (Core) and ISO 18629– SPAR - Shared Planning and Activity Representation

I-X Resources – I-X 4.6 (25-Aug-2010) – I-X Applications – I-X UAV-MP App for UAV Mission Planning (2003)– I-Globe app (I-X/AGlobe) planning and simulated

execution for mobile vehicles operation (2007)

I-Room app - a virtual space for intelligent applications

11



Domain FeaturesDomain Features

problem formulation: planning domain + problem define search space; essential for efficient planning

idea: explicitly represent (redundant) domain features to ensure consistency of the representation

overview (domain features):– static vs. fluent relations– domain types– reversible actions– inconsistent effects

PDDL: contains types (optional); other features not supported

12



Static vs. Fluent Static vs. Fluent RelationsRelations

example: DWR domain(:predicates

:static (adjacent ?l1 ?l2 - location) :fluent (at ?r - robot ?l - location)

etc.(:action move

:parameters (?r - robot ?from ?to - location):precondition (and

(adjacent ?from ?to) (at ?r ?from)(not (occupied ?to)))

:effect (and (at ?r ?to) (not (occupied ?from))(occupied ?to) (not (at ?r ?from)) ))

domain validation:– static relations: must not appear in effects

– fluent relations: should appear somewhere in effects

13



Domain Types: ExampleDomain Types: Example

example: (:predicates

:static (adjacent ?l1 ?l2 - location) :fluent (at ?r - robot ?l - location)

etc.(:action move

:parameters (?r - robot ?from ?to - location):precondition (and

(adjacent ?from ?to) (at ?r ?from)(not (occupied ?to)))

:effect (and (at ?r ?to) (not (occupied ?from))(occupied ?to) (not (at ?r ?from)) ))

domain validation:– derive type system from operator

specification

– compare declared types to derived types

14



Derived Types (O-types)Derived Types (O-types)

type names: <N,i> (predicate/operator arg. position) O-type: set of type names such that <N1,i1> O <N2,i2> iff

variable v: v is i1th parameter of operator N1 which has a precondition or effect with predicate N2 and v as i2th argument, or

– [symmetric case for operator N2 and predicate N1], or

– [transitivity of O relation]

Proposition: Let (si,g,O) be a STRIPS planning problem and let (C, T, ) be the O-type partition derived from this problem. Then every state that is reachable from the initial state si is type consistent.

Theorem:Let (si,g,O) be a STRIPS planning problem and let (C, T, ) be the O-type partition derived from this problem. If two constants c1 and c2 have the same type (c1) = (c2) then they must have the same type in every type partition that creates a type consistent search space.

15



Derived Types: Derived Types: AlgorithmAlgorithm

for every operatorfor every precondition and effect

for every argument variable v

let tpi be the O-type of the predicate/argument position

let tv be the O-type of the variable

tpi undefined and tv undefined:

create new O-type; assign to tpi and tv

tpi defined and tv undefined:

assign tpi to tv

tpi undefined and tv defined:

assign tv and tpi

tpi defined and tv defined:

merge O-types of tpi and tv

16



Domain Types: Domain Types: EvaluationEvaluation

DWR domain (types defined)– algorithm output:

» type: [loaded-0, unloaded-0, at-0]» type: [attached-0, top-1, in-1]» type: [occupied-0, attached-1, belong-1,

adjacent-1, adjacent-0, at-1]» type: [belong-0, holding-0, empty-0]» type: [loaded-1, holding-1, on-1, on-0, in-0,

top-0]– corresponds to five types declared in PDDL

specification other domains (from IPC, no types defined)

– movie, gripper, logistics, mystery, mprime and grid

– algorithm derives between 3 and 5 types– “manual” evaluation appears consistent

17



OpenVCE: Virtual SpaceOpenVCE: Virtual Space

I‑Room: an environment designed for intelligent interaction

realizations in Second Life® and OpenSim

avatars in surrogate reality: intuitive grounding in a persistent space

compelling sense of shared presence

intelligent systems and tool support for meetings and collaborative activities: I-X and <I-N-C-A>

18



Collaborative SOP Collaborative SOP DevelopmentDevelopment

idea: use wiki as tool for collaborative SOP development– MediaWiki: open source, scalable, robust

tool– SOP extension: supports structured SOPs

creating an SOP on the wiki:– write the unstructured SOP as free text (or

use existing book)– divide into articles; one

method/refinement per article– mark up articles using SOP extension tags:

» {{#objective:…}} the task accomplished by the method; one per method

» {{#subtask:…}} task breakdown; tasks that need to be accomplished (in sequence) to implement the method

19



Distributed Use of SOPsDistributed Use of SOPs

idea: export formal SOP (based on tags) to task support tool– I-X: Intelligent Distributed To-Do List or

– web-based tool with similar functionality

20



What’s Next?What’s Next?

Making domain analysis more accessible with a GUI (in progress)

Work package 1:– Analysis of collaborator applications

» What is the problem?

» What data is available?

» What degree of autonomy / human intervention is expected?

– State of the art

Work package 2: Simulation Work package 3: Representation

1 hedlamp kick-off meeting schlumberger cambridge, 3/4/2012 hedlamp (huddersfield + edinburgh -...

Documents