18
AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics, Email: [email protected]

AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

  • View
    217

  • Download
    0

Embed Size (px)

Citation preview

Page 1: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS

Lee McCluskey

Professor of Software Technology,

Dept of Informatics,

Email: [email protected]

Page 2: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Outline

To give you an idea, using NASA’s work on autonomy as an example, what is the function of “automated planning software”

To give you an idea of the use of GIPO, a tool to help people utilise automated planning software

Page 3: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

What is Autonomy? WordNet Definitions:

S: (n) autonomy, liberty (immunity from arbitrary exercise of authority: political independence)

S: (n) autonomy, self-direction, self-reliance, self-sufficiency (personal independence)

Page 4: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

What is Autonomy? My Definition of Autonomy as it applies to Software..

Autonomous software ..

Deliberates and Acts: Can choose and perform its own actions / activities

Senses: Has an “awareness” of its own environment and internal state

Accepts Commands: Is controlled by being given tasks to perform or goals to achieve

(+ lots of other stuff to do with Learning and Adaptation .. left out here..)

Page 5: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

What is not Autonomy?

Honda’s biped “Asimo” – Asimo is the best robot in the world according to http://www.robot.org.uk

It can walk and recognise some objects. It has very good movement but not much intelligence!

Picture from this website.

Page 6: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Example of an Intelligent Autonomous Agent - Deep Space 1

1998 – 1999

NASA demonstrated that planning software called “Remote Agent” could be used to manage a spacecraft the Deep Space 1 .

The spacecraft was controlled for a portion of its journey with little human intervention.

Page 7: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Example of an Intelligent Autonomous Agent - The 2007 Mars Exploration Rover

pictures from NASA-Ames Research Website

COLLECT A ROCK SAMPLE AT CRATER 3 THEN GET BACK TO

THE MOTHER SHIP BEFORE SUNSET..

Aye Aye Captain !

Millions of Miles

Page 8: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Necessary equipment for IAA ..?communicationmechanism with its controller sensing mechanism

motor control

WITHOUT PLANNINGCAPABILITY

Page 9: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Necessary equipment for IAA ..?

a rich dynamic store (database) representing the outside world – facts, events, actions, processes ..

a program to reason with the database and form plans of action to satisfy the controller’s requests, and schedules to carry out the plans

communicationmechanism with its controller sensing mechanism

motor control

WITH PLANNINGCAPABILITY

Page 10: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Automated Planning and Scheduling ..

a rich dynamic store (database) representing the outside world – facts, events, actions, processes ..

a program to reason with the database and form plans of action to satisfy the controller’s requests, and schedules to carry out the plans

“KNOWLEDGE” BASEPLANNING – assembling orderingsof actions in order to achieve goals or tasks

SCHEDULING – assigning actions to fixed times andresources

Page 11: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Automated Planning and Scheduling .. Challenges in Knowledge Acquisition

a rich dynamic store (database) representing the outside world – facts, events, actions, processes ..

“KNOWLEDGE” BASEFormulating this kind of knowledge is VERY HARD!

Formulating knowledge of continuous processes and uncertain outcomes of actions, in particular, is still a great challenge for AI

Page 12: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Automated Planning and Scheduling .. Challenges in Automated Planning

a program to reason with the database and form plans of action to satisfy the controller’s requests, and schedules to carry out the plans

PLANNING – assembling orderingsof actions in order to achieve goals or tasks

SCHEDULING – assigning actions to fixed times andresources

Automated Planning is VERY HARD! Many automated planning programs exist but they make simplifying assumptions about the outside world.

Page 13: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Some Other Applications of Automated Planning and Scheduling..

Satellite mission planning and scheduling Spacecraft Assembly Automated image processing Planning for military air campaigns Planning for Crisis Management Brewery production-line scheduling Generation of control programmes for industrial plant Oil Spill Response Planning Ship Building using a shipyard scheduling optimisation system Aircraft crew scheduling Chemical Plant control Autonomous Vehicles control Workflow/Workforce management Web Service Composition

Page 14: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

GIPO

- GIPO is a software tool that addresses some of the Challenges in Knowledge Acquisition described above: it is a GUI for helping building up the knowledge base for use in a planning application.

- GIPO won the best tools award at the First International Competition in Knowledge Engineering for AI Planning at Monterey, California in 1995.

- Scientists at NASA-Ames have used GIPO, and are keen to see it adapted for use with their knowledge representations. A future version of GIPO could well be used in future space missions…

Page 15: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

The “Briefcase” World – a simple logistics example

Locations

Things

Bags

The Mars rover has been given a day job!It has to produce plans to transport mundane things from home to work to town ….

Page 16: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

putin(briefcase,home,cake)move(briefcase,home,work)takeout(briefcase,work,cake)move(briefcase,work,town)putin(briefcase,town,letter)move(briefcase,town,work)putin(briefcase,work,tools)move(briefcase,work,home)takeout(briefcase,home,tools)

INITIAL STATE

GOAL

PLAN

The planning function requires a planner a knowledge base and a specification of the tasks

Page 17: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

GIPO Demonstration

- To attempt to do something NO-ONE has ever done before during a (public) lecture …. Create a planning application form scratch ….

- Resources:Planning and Scheduling programs:

http://scom.hud.ac.uk/planet/repository/GIPO:

http://scom.hud.ac.uk/planform/gipo/UK Planning Special Interest Group:

http://planning.cis.strath.ac.uk/plansig/

Page 18: AUTOMATED PLANNING SOFTWARE: ENABLING TECHNOLOGY FOR (Intelligent) AUTONOMOUS AGENTS Lee McCluskey Professor of Software Technology, Dept of Informatics,

Artform Research Group

Summary

- There are useful applications of Automated Planning and Scheduling in Space, Logistics, Crisis Management etc. But current applications tend to embody simplified views of the world (for example, they assume no uncertainty, or they assume all processes/actions involve discrete change)

- NASA are currently one of the leaders in this area. Scientists at NASA-Ames have used GIPO, and are keen to see it adapted for use with their knowledge representations.

- GIPO is a knowledge acquisition tool which helps people encode knowledge that can be used with automated planning programs