Transcript AIAI Presentation

___________________________________________________
Intelligent Planning and Collaborative Systems
for Emergency Response
UNIQUE AIAI
RESOURCES
http://i-x.info
http://i-rescue.org
•More than 20 years of excellence in applied Artificial Intelligence
•World-leading AI planning research and technical team
•World-leading knowledge modelling and representation resources and staff
•O-Plan: Multi-Perspective Planning Architecture and Planning Web Service
•I-X: Issue Handling Planning and Collaboration Architecture
•<I-N-C-A>: Knowledge Elicitation, Encoding, Modelling, Representation, and
Management
•I-X commercialisation through Scottish Enterprise Proof-of-Concept Award:
IM-PACs
This briefing is available in http://www.aiai.ed.ac.uk/~bat/jp/
Edinburgh AI Planners in Productive Use
http://www.aiai.ed.ac.uk/project/plan/
2
Intelligent Messaging, Planning and Collaboration Systems for
Emergency Response
GOALS
TASKS
COMMUNICATION
DRAFT RESPONSE PLANS:
MULTIPLE COURSES OF
ACTION
Effects-Oriented Planning
Planning System
Issues
EVENT
Choose (IH)
I Issues or Implied Nodes
Constraints
Do (IH)
Node
N
Constraints
Constraints
Detailed
Propagate
C
Space
of
Legitimate
Solutions
IH=Issue Handler
A=Annotations
Constraints
Constraints
(Agent Functional Capability)
KNOWLEDGE
BASE
Shared Task and
Activity Model
KNOWLEDGE
MODELLING
I-X: Issue
Handling and
Task Support
Architecture
Knowledge about places, people, processes, infrastructure,
connectivity, response capabilities, and meta-knowledge
O-Plan/I-Plan:
Multi-Perspective
Planning
AIAI TECHNOLOGIES
COORDINATED RESPONSE
EVENT
AUTOMATED
REASONING
DECISION
MAKING
RESPONSE TEAM
<I-N-C-A>: Knowledge
Elicitation, Encoding,
Modelling,
Representation, and
Management
A More Collaborative & Dynamic
Planning and Execution Framework
 Human relatable and presentable objectives, issues, sensemaking, advice, multiple options, argumentation, discussions
and outline plans for higher levels
 Detailed planners, search engines, constraint solvers, analyzers
and simulators act as services in this framework in an
understandable way to provide feasibility checks, detailed
constraints and guidance
 Sharing of processes and information about process products
between humans and systems
 Current status, context and environment sensitivity
 Links between informal/unstructured sense-making and
discussion and more structured planning, methods for
optimisation and decision support
4
I-X
Multi-Agency Emergency Response Planning,
Execution, and Task-Oriented Communications
Collaboration and
Communication
Central
Authorities
Command
Centre
Emergency
Responders
Isolated
Personnel
5
<I-N-C-A> Framework
 Common conceptual basis for sharing information on processes and
process products
 Shared, intelligible to humans and machines, easily communicated,
formal or informal and extendible
 Set of restrictions on things of interest:
• I
•N
•C
•A
Issues
Nodes
Constraints
Annotations
e.g. what to do? How to do it?
e.g. include activities or product parts
e.g. state, time, spatial, resource, …
e.g. rationale, provenance, reports, …
 Shared collaborative processes to manipulate these:
• Issue-based sense-making (e.g. gIBIS, 7 issue types)
• Activity Planning and Execution (e.g. mixed-initiative planning)
• Constraint Satisfaction (e.g. AI and OR methods, simulation)
• Note making, rationale capture, logging, reporting, etc.
 Maintain state of current status, models and knowledge
 I-X Process Panels (I-P2) use representation and reasoning together
with state to present current, context sensitive, options for action
Mixed-initiative collaboration model of mutually constraining things
6
I-X Approach
 The I-X approach involves the use of shared models for taskdirected communication between human and computer agents
 I-X system or agent has two cycles:
• Handle Issues
• Manage Domain Constraints
 I-X system or agent carries out a (perhaps dynamically
determined) process which leads to the production of (one or
more alternative options for) a “product”
 I-X system or agent views the synthesised artifact as being
represented by a set of constraints on the space of all possible
artifacts in the application domain
7
<I-N-C-A>
Product Model
I
Issues or Implied
Constraints
Issues
N
Node
Constraints
Nodes
C
Detailed
Constraints
Constraints
Space of Legitimate Product Models
A
Annotations
8
I-X and <I-N-C-A>
Product Model
I
Issues or Implied
Constraints
Issues
N
Node
Constraints
Nodes
C
Detailed
Constraints
Constraints
Space of Legitimate Product Models
A
Choose (IH)
Do (IH)
Propagate
Constraints
IH=Issue Handler
(Agent Functional Capability)
Annotations
9
I-P2 aim is a Planning, Workflow and
Task Messaging “Catch All”
 Can take ANY requirement to:
•
•
•
•
Handle an issue
Perform an activity
Respect a constraint
Note an annotation
 Deals with these via:
•
•
•
•
•
Manual activity
Internal capabilities
External capabilities
Reroute or delegate to other panels or agents
Plan and execute a composite of these capabilities (I-Plan)
 Receives reports and interprets them to:
• Understand current status of issues, activities and constraints
• Understand current world state, especially status of process products
• Help user control the situation
 Copes with partial knowledge of processes and organisations
10
Anatomy of an
I-X Process Panel
11
I-X Process Panel and Related Tools
Domain Editor
Process Panel
Messenger
Map Tool
I-Plan
12
I-Space and I-World
13
14
Safety and Companion Robots
15
e-Response Vision
 The creation and use of task-centric virtual organisations
involving people, government and non-governmental
organisations, automated systems, grid and web services
working alongside intelligent robotic, vehicle, building and
environmental systems to respond to very dynamic events on
scales from local to global.
 Multi-level emergency response and aid systems
 Personal, vehicle, home, organisation, district, regional, national,
international
 Backbone for progressively more comprehensive aid and emergency
response
 Also used for aid-orientated commercial services
 Robust, secure, resilient, distributed system of systems
 Advanced knowledge and collaboration technologies
 Low cost, pervasive sensors, computing and comms.
 Changes in building codes, regulations and practices
16
e-Response Relevant Technologies

Sensors and Information Gathering
•
•
•
•
•
•
•

Emergency Response Capabilities and Availability
•
•
•

local versus centralized decision making and control
mobile and survivable systems
human and automated adjustable autonomy mixed-initiative decision making
mixed-initiative, multi-agent planning and control
trust, security
Common Operating Methods
•
•
•
•
•

robust multi-modal communications
matching needs, brokering and "trading" systems
agent technology for enactment, monitoring and control
Hierarchical, distributed, large scale systems
•
•
•
•
•

sensor facilities, large-scale sensor grids
human and photographic intelligence gathering
information and knowledge validation and error reduction
semantic web and meta-knowledge
simulation and prediction
data interpretation
identification of "need"
shared information and knowledge bases
Shared standards and interlingua
shared human scale self help web sites and collaboration aids
shared standard operating procedures at levels from local to international
standards for signs, warnings, etc.
Public Education
•
•
•
publicity materials
self help aids
public training
17
FireGrid Technologies
http://firegrid.org
Tens of Thousands of
Sensors & Monitors
Emergency
Responders
Knowledge Systems,
Planning & Control
Super-real-time
Simulation
Computational Grid
Maps,
Models,
Scenarios
18
FireGrid Overview
http://firegrid.org
 Mission statement:
- …to establish a cross-disciplinary collaborative community to pursue
fundamental research for developing real time emergency response systems
using the Grid…
- Initial domain is fire emergencies.
 Challenges:
- Sensing: instantaneous and continuous relay of data from emergency location to
response system via the Grid.
- Modelling: model the evolution of fire and impact on building, and relate this to
intervention alternatives and evacuation strategies.
- Forecast: all simulations, analyses and communications done in ‘super real-time’.
- Response: effective co-ordination of response with intelligent decision-support system.
- Feedback: continuously update simulations, predictions and response using latest data
from sensors and responders.
 Status:
- DTI/University of Edinburgh/Industry-funded project, total value: £2.23M, start
date: 1st March 2006.
- Modelling Emergencies in Real-Time from Sensor Input (MERSI) research project
at initial (EPSRC) proposal stage.
19
The FireGrid Cluster
 Universities and Colleges:
http://firegrid.org
- University of Edinburgh; Imperial College London; Queen Mary, University of
London; The Fire Service College, UK; Institute of High Performance Computing,
Singapore; TU Delft, The Netherlands; IHMC Florida
 National Research Laboratories:
- National e-Science Centre, UK; Health and Safety Laboratory, UK; NIST, USA;
Major Accident Prevention Division, IRSN, France; TNO Building and Construction
Research, The Netherlands.
 Computational Software and Sensing Technology Companies:
- Vision Systems (Europe) Ltd.; ABAQUS UK Ltd.; ANSYS Europe Ltd.; Integrated
Environment Solutions Ltd.
 Engineering and Technology Consultancy Companies:
- Arup Fire; BRE Building Research Establishment Ltd.
 Emergency Planning and Response:
- Fire Research Division, Office of the Deputy Prime Minister, UK; London Fire and
Emergency Planning Authority; Lothian and Borders Fire Brigade, Edinburgh;
Greater Manchester County Fire and Rescue Service.
20
Cycle 20
Search and Rescue
Command Centre
Ambulance
Centre
Fire
Station
Blocked Roads
Roads
Police
Office
Buildings
Cycle 200
RoboCup Rescue Simulator
Simulates the Kobe earthquake
Sends sensorial information to
agents, receiving back action
commands
I-X Agents
Divided in three hierarchical
decision-making levels
Support ideas such as activity
oriented planning, coordination and
knowledge sharing
Interaction I-X to Kobe Simulator
Information from RCRS to I-X is
converted to the <I-N-C-A> format
Ambulance Team
Fire Brigade
Police Force
Adapted from H. Kitano and S. Tadokoro, RoboCup Rescue A Grand Challenge
for Multiagent and Intelligent Systems, AI Magazine, Spring, 2001.
21
http://www.capwin.org
Galileo
http://www.esa.int/navigation/galileo/
More Information
• www.aiai.ed.ac.uk/project/plan/
• www.aiai.ed.ac.uk/project/ix/
• i-rescue.org
• i-x.info
• i-c2.com
24
Prof. Austin Tate
• Technical Director, Artificial Intelligence Applications Institute
• Professor of Knowledge-Based Systems, University of Edinburgh
• Fellow of the Royal Society of Edinburgh (Scotland's National
Academy), Fellow of the American Association for AI, Fellow of the
British Computer Society, Fellow of the International Workflow
Management Coalition, and a member of the editorial board of a
number AI journals.
.
• His internationally sponsored research work involves advanced
knowledge and planning technologies, especially for use in emergency
response and search and rescue.
25
Spare Slides
 Spare Slides
26
High Level Planning and Activity Management
Sensors, User Inputs, E-mail, External Influences
Sub-plan Library
HTN Planning
&
<I-N-C-A>
Diary
Behaviours: Preprogramed, Situation-Response, Reactive
27
HTN Planning
Activity Composition
Plan Library
A2 Refinement
S1
S2
“Initial” Plan
“Final” Plan
A2.1
A2
A2.2
Refine
A4
A1
A3
A5
A4
A1
A5
A3
Introduce activities to achieve preconditions
Resolve interactions between conditions and effects
Handle constraints (e.g. world state, resource, spatial, etc.)
28
HTN Planning
Initial Plan Stated as “Goals”
Plan Library
Ax Refinement
S1
S2
P
“Initial” Plan
“Refined” Plan
P
P
Refine
Q
A1.1
A1.2
Q
Initial Plan can be any combination of Activities and Constraints
29
Some Planning Features
 Expansion of a high level abstract plan into greater detail
where necessary.
 High level ‘chunks’ of procedural knowledge (Standard
Operating Procedures, Best Practice Processes, Tactics
Techniques and Procedures, etc.) at a human scale - typically
5-8 actions - can be manipulated within the system.
 Ability to establish that a feasible plan exists, perhaps for a
range of assumptions about the situation, while retaining a
high level overview.
 Analysis of potential interactions as plans are expanded or
developed.
 Identification of problems, flaws and issues with the plan.
 Deliberative establishment of a space of alternative options,
perhaps based on different assumptions about the situation
involved, of especial use ahead of time, in training and
rehearsal, and to those unfamiliar with the situation or utilising
novel equipment.
30
More Planning Features
 Monitoring of the execution of events as they are expected to
happen within the plan, watching for deviations that indicate a
necessity to re-plan (often ahead of this becoming a serious
problem).
 Represent the dynamic state of the world at points in the plan
and use this for ‘mental simulation’ of the execution of the
plan.
 Pruning of choices according to given requirements or
constraints.
 Situation dependent option filtering (sometime reducing the
choices normally open to one ‘obvious’ one.
 Satisficing search to find the first suitable plan that meets the
essential criteria.
 Heuristic evaluation and prioritisation of multiple possible
choices within the constrained search space.
 Uniform use of a common plan representation with embedded
rationale to improve plan quality, shared understanding, etc.
31
Human Approach
 Previous slides describe aspects of problem solving
behaviour observed in expert humans working in
unusual or crisis situations.
 Gary Klein, “Sources of Power”, MIT Press, 1999.
 But they also describe the hierarchical and mixed
initiative approach to planning in AI developed over
the last 25 years.
32
Compendium
http://www.compendiuminstitute.com
Compendium
http://www.compendiuminstitute.com
35
<I-N-C-A> Ontology
Issues
Outstanding questions, problems or requirements (gIBIS)
Nodes
E.g. activities in a process or parts in a physical product
Constraints
Critical Constraints (shared across multiple components)
Auxiliary Constraints (localised to a single component)
Annotations
E.g. decision rationale and other notes
36