SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ... · SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (2024)

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (1)

RL-TR-92-2 1 7 Final Technical Report August 1992

SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE 0-PLAN2 PROJECT

University of Edinburgh

Austin Tate, Brian Drabble, Richard Kirby

APPROVED FOR PUBL/C RELEASE; D/STR/BUl/lr/ON UNL/MT€D

Rome Laboratory Air Force Systems Command

Griffiss Air Force Base, NY 13441 -5700

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (2)

This r e p o r t has been reviewed by t h e Rome Laboratory Publ ic A f f a i r s Of f i ce (PA) and is r e l e a s a b l e t o the Nat ional Technical Information Service (NTIS). f o r e i g n n a t i o n s .

A t NTIS i t w i l l be r e l e a s a b l e t o the gene ra l p u b l i c , i nc lud ing

RL-TR-92-217 has been reviewed and i s approved f o r p u b l i c a t i o n .

APPROVED :

NORTHRUP FOWLER %I1 Pro j e c t Engineer

, FOR THE COMMANDER:

JOHN A . GRANIERO Chief S c i e n t i s t Command, Control & Communications D i r e c t o r a t e

r

I f your addres s has changed o r i f you wish t o be removed from t h e Rome Laboratory ma i l ing l i s t , o r i f t h e addressee i s no longe r employed by your o rgan iza t ion , p l e a s e n o t i f y RL (C3C ) G r i f f i s s AFB NY 13441-5700. This w i l l assist u s i n maintaining a c u r r e n t ma i l ing l i s t .

D o n o t r e t u r n cop ie s of t h i s r e p o r t u n l e s s c o n t r a c t u a l o b l i g a t i o n s o r n o t i c e s on a s p e c i f i c document r e q u i r e t h a t i t be r e tu rned .

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (3)

REPORT DOCUMENTATION PAGE 1 PhZ$pA?Ed0188

. AGENCY USE ONLY (Leave Blank) 2 REPORT DATE 3. REPORT TYPE AND DATES COVERED August 1992 F i n a l Jun 89 - May 92

. TITLE AND SUBTITLE SPACECRAFT COMMAND & CONTROL USING AT TECHNIQUES - THE 0-PLAN2 PROJECT

PUNNING

I. AUMOR(S) Aust in T a t e , Brian Drabble , Richard Kirby

'. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) Art i f i c i a l I n t e l l i g e n c e A p p l i c a t i o n s I n s t i t u t e U n i v e r s i t y of Edinburgh 80 South Bridge Edinburgh EH1 l H N , UK

5. FUNDING NUMBERS C - F49620-89-C-0081 PE - 62702F PR - 5581 TA - 27 Flu - 44

8..PERFORMING ORGANIZATION REPORT NUMBER

AIAI-TR-109

4. SUBJECT TERMS

Planning , A r t i f i c i a l I n t e l l i g e n c e

7. SECURTP/ ClASSlFlCATlON 18. SECURIJY ClASSlFlCATlON 19. SECURIJY CLASSIFICATION OF REPORT OF THIS PAGE OF ABSTR&CT UNCLASSIFIED UNCLASSIFIED UNCLA SIFIED

I. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) Rome Labora tory (C3C) 525 Brooks Road Gr i f f iss AFB NY 13441-450-5

15 NUMBER F PAGES %O

16 PRICE CODE

20. LIMITATION OF ABSTRAC

UL

10. SPONSORING/MONITORING AGENCY REPORT NUMBER

RL-TR-92-217

1. SUPPLEMENTARY NOTES

Rome Laboratory P r o j e c t Engineer: Northrup Fowler IIIf C3Cf (315) 330-3011

2a. DISTRIBUTlON/AVAIIABIW STATEMENT

Approved f o r p u b l i c r e l e a s e ; d i s t r i b u t i o n u n l i m i t e d .

11%. DISTRIBUTION CODE

I

3. ABSTRACTW- 200 wads) 0-Plan2 ( t h e ODen Planning A r c h i t e c t u r e ) p r o v i d e s a g e n e r i c domain independent - computat ional a r c h i t e c t u r e s u i t a b l e f o r command, p lanning and e x e c u t i o n a p p l i c a t i o n s . v i s i o n of a modular and f l e x i b l e p lanning and c o n t r o l system i n c o r p o r a t i n g a r t i f i c i a l i n t e l l i g e n c e methods.

This r e p o r t d e s c r i b e s t h e 0-Plan2 a g e n t o r i e n t e d a r c h i t e c t u r e and d e s c r i b e s t h e communication which t a k e s p l a c e between p lanning and e x e c u t i o n moni tor ing a g e n t s b u i l t upon t h e a r c h i t e c t u r e . w i t h i n t e r n a l and e x t e r n a l i n t e r f a c e s p e c i f i c a t i o n s t h a t form a p a r t o f t h e d e s i g n . A d e s c r i p t i o n of t h e p r o t o t y p e implementat ion of 0-Plan2 i s i n c l u d e d and t h e r e p o r t d e s c r i b e s an a p p l i c a t i o n OS 0-Plan2 t o t h e g e n e r a t i o n of on-board commands f o r a s imple, bu t r e a l i s t i c , s p a c e c r a f t .

The main c o n t r i b u t i o n of t h e 0-Plan2 r e s e a r c h has been a complete

S e p a r a t e modules of such a system are i d e n t i f i e d a long

, I I SN 754001-2805500 Standard Form 298 Rev 2-8'

Presalbed by ANSI &td 239- 298-1 02

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (4)

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (5)

Acknowledgements

The O-Plan project began in 198 1. Since t h a t time the following people have participated: Colin Bell, Ken Ciirrie, Jeff Dalton. Roberto Desiruone, Brian Drabble, Mark Drulnniond, Anja IIaman, Ken Johnson, Richaid Kirby. .\rthur Seatoti. Judith Secker, Aiistiii Tate aiid R i cli a r d Tobiii.

Prior to 1984, work on Interplan ( I 972-.4) aiid N;)nliii ( 1975-6) was fuiiclcd hy the U K Science and Engineering Research Couiicil.

From 1984 to 1988, the O-Plan projcc t \va5 funded 1)). the Science and Engineering Research <hiiicil on grant numbers GR/C/591;s and G n / ~ / 5 8 9 8 7 ( U I < Alvey Programme project number 11<~8/1.51). The work was also snpportcd by a fcllon.~liip from SD-scicoii for Anstin Tate from 1084 to 1985.

From 1989 to 1992, the O-Plan2 projcct lrah I,w.n support(v1 11y the us Air Force Roiiie Labo- ratory through the Air Force Office of Sc icnt ific Research (.\I’osR) and their Europeaii Office of Acrospace Research and Developnieiil 1)). cont rac t n i i i ~ i l w r I’ 1962O-S9-C‘OOSl ( ~ 0 ~ ~ ~ / 8 8 - 0 0 4 4 ) monitored by Nortlirup Fowler 111 at t l r c I

Additional resoiirces for the O-Plan ant1 O-l’lan2 project5 have been proviclcd by the Artificial Intelligence Applicatioiis Institnlc 111 ioiigli t Iro 1 ; t ’ R o I ’ \ (Edinburgh Vniversity Research on Planning Architectures) Institute ( l (~v(~lopir t (~i i1 1)iojc.c t .

Since 1989, research on scheduling applicat io115 of thr O-Plan architecture has been funded 1)y Hitachi Europe Ltd. Froni 1989 t o iSc)’L. 1 l i ~ S c i w c o and Engineering Research Couii- cil (grant number GR/F36%5 - t l l i I i i fo i .~ i~a t i o i i ~ i r g ~ i r ~ ( ~ i ~ i i i g Directorate project iiuinber I E D 4/1/1320) has funded a collahorat i\-v project \vit I r ] ( * I , . Impc>riaI CoIIcge and other partners in which the O-Plan architecture i i h i i rg r i d t o guide ~ h c tle\ign and development of a planner with a flexible temporal logic reprwiiit a t io11 of t I r c plan state. -4 iiiiiii1)er of other research and development contracts placed wi th . ~ I . \ I Ira\(’ I c t l lo r c v a i t h 1)rogrehs on the O-Plan prototype.

it’ Roinc 1,al)oratory.

iii

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (6)

Abbreviations

ADS Associated Data Structure - the level of dii ta htriictiire in 0-Plan2 a t which a plan is represented. This is “associatctl” witli i l l1 iititlci~lying l’iine Phn t Network ( T P N ).

(s

iv

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (7)

V

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (8)

Contents

Acknowledgements

Abbreviations

Contents

i ii c

iv

vi

1

1.1 Project A i m 1

1 Summary

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.2 Project Acltirvements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

# 3 2 Introduction

5

3.1 0-Pla.nl 5

3 History and Technical Influences

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 0-Plan2 . . . . . . . . . . . 6 3 .:3 Char ac teri sa.t i on of 0 - PI il 11 2 ’7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9 Communication in Command, Planning and Control

4.1 TheScenario 9

4.2

4.3

4

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . IJse of Dependencies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

A Common Representatioil f‘or ( ‘ o i i ) I I I I I II i c i i 1 ioii Iwt IVCCII :!gc~rts . . . . . . . . . . 10

12

5.1 P lans t a t e s 12

5.1.1 Task Formalism ( T F ) 12

ri.1.2 12

5.2 Plan Patches 13

5 Representing and Communicating Plans

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Plan Flaws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

.5.3 Plan Patch r \ t t a c l ~ m e ~ ~ t I’oitit\ . . . . . . . . . . . . . 1-1- . . . . . . . . . . . . . . .

.5.4 Incremental Plan States 14 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .>..I r r Plan Transartions 1.5

16 G Managing Concurrent Conip ut a t lolls *

vi

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (9)

6.1 Choice Ordering hIcchaiii~ii~s i n O-l’laii 1 . . . . . . . . . . . . . . . . . . . . . . . 16

6.1.1 Building 111) Tiiforutilt i o i i i i i i\~i . \g(’lltlii R ( ~ o ~ . t l . . . . . . . . . . . . . . . 16

6.1.2 Granularity of Iitto\vlcdgc Soiirceh . . . . . . . . . . . . . . . . . . . . . . l G

6.1.3 Priority of Proces~ing .\p, endd 1‘iit ri(\\ . . . . . . . . . . . . . . . . . . . . 17

6.2.1 Iinowledge Source Sti lgch . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

6.2.2 Knowledge Soiircc Triggcrh . . . . . . . . . . . . . . . . . . . . . . . . . . 18

6.2.3 C‘ompound Agenda I < n t rics . . . . . . . . . . . . . . . . . . . . . . . . . . 19

6.2.4 Controller Priorit io> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

6.2 Choice Ordering hfecliani\ttis in O-l’iilli2 . . . . . . . . . . . . . . . . . . . . . . . 17

7 0-Plan2 Architecture 20

7.1 Domain Tiiforination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

7.2 PlariState . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

7.3 Tinowledge Soitrres . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

T . I Support Iblotlrlles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

7.5 (“ontrollcr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

7.(j Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

7.6.1 rinolvlcdge SoIlrc1~h . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

7.6.2 Controller Strategic> . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

I . 1 Process Stritcture of the. 0-1’1~112 l ~ r i ~ ) l ( ~ i i i ( ~ ~ r t i i t i o ~ ~ . . . . . . . . . . . . . . . . . . 26

7.S Processing Cycle in tiic ( ‘ t i l i v i i i O - l ) I~ i i i 2 S,i \ ~ c i i i . . . . . . . . . . . . . . . . . . 28

- -

8 0-Plan2 Planner 30

S.l P1a.n State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

8.1.1 Plan Net. worlc - AI)^ ; i i i t l TI ’S . . . . . . . . . . . . . . . . . . . . . . . . . 30

5.1.2 ‘J’O\lIC atid GOS‘l’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

S.l.:3 l)lili1 Slate \ ~ 7 i \ l ~ i i i 1 ) l ( ~ ~ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

X . l .. 1 Ilcsoiirce ITtilisiitioii . i..il)l(% . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1

s.1.5 Ag(wda . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

S.2 Planning I<no\vlcvlgc~ S o ~ i i ~ c ~ s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . :32

8.13 lisp of (‘oirstraint . L f i i l l i \ ~ ( ~ ~ ~ ~ 1 0 .\ l.iiiitiiiii I’IiIiI lliIbt~I~iiiti(~ii . . . . . . . . . . . . . 3 3

8 ..I. I Tiiiic Point Y ( ~ t \ ~ o l . l < l i i \ l i i i ~ ~ ~ l ~ ( . I . I ) s \I ) . . . . . . . . . . . . . . . . . . . . 33

vii

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (10)

8.3.2 T O M E / G O ~ T h h a g c t ( K V ) . . . . . . . . . . . . . . . . . . . . . . . . . :?6

8.3.3 Resoiirce TJtilisation llanagc~iiicnt (1tt \ I ) . . . . . . . . . . . . . . . . . . . 138 8.3.4 Plan State Variahlc5 hl;luagrr ( r J \ \ . l l ) . . . . . . . . . . . . . . . . . . . . 36

8.4 Support Mechanisms i n @Plan2 . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

8 ..5 Alternatives Manager . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

8.6 Implementat. ion as Separatc I’~m-e\\c\ . . . . . . . . . . . . . . . . . . . . . . . . 38 c

9 0-Plan2 Job Assigner 40

10 0-Plan2 Execution System 41

11 0-Plan2 User Interface 43

11.1 Planner IJser Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

11.2 System Developer Interfaw . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -15

11.3 0-Plan2 User Roles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

11.3.1 Domain Expert Rolc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

11.3.2 Doinaiii Specialist Rol(1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

11.3.3 Command User Rolc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

11.3.4 Planner User Role . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

11.3.5 Execution Sj.stei i i \\.,it( l i / S l o ( l i l j Ilol(> . . . . . . . . . . . . . . . . . . . . 1;

11.3.6 System Deve1opc)r Ilol(1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

11.3.7 Vser Support to ( ‘ o l r l t ~ ) l l c i Holv . . . . . . . . . . . . . . . . . . . . . . . 47

1 1.3.8 77ser SU pport to -1 1 I VI’ I I i i I i 1.0 1 I rl I I r ~ g ~ I . . . . . . . . . . . . . . . . . . . . .

11.3.9 TTser as Systein 13iiildv1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47 ~

12 Performance Issues and Instrument a t’ 1 0 1 1 48

12.1 Architecture Perfornlancc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . -15

12.2 Constraint hfanager ant1 Stip1)011 l l o u t i t r c ~ I ’ (~~for i t i~ i r~c.c. . . . . . . . . . . . . . . 49

1’2.13 hlonitors and Tnstruinrwt (11 io t i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

13 Modularity. Interfaces aiicl Protocols 50

13.1 C‘omponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

13.2 Support hlotlules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

13.3 Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (11)

13.4 Interiial Snppovt Fa.cilitics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.4.1 Iiiiowledge S o ~ r c c T : Y ~ I I I C \ V : O Y I ~ ( K S F ) . . . . . . . . . . . . . . . . . . . . . 52

13.4.2 Agenda Trigger Language . . . . . . . . . . . . . . . . . . . . . . . . . . . 13.4.3 Controller Priority 1,angua.ge . . . . . . . . . . . . . . . . . . . . . . . . . ,54

13..5 External 1iiterfa.ces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

51

53

14 Spacecraft Coininand and Control Application 5 5

15 Related Projects 58

16 Future Plans for 0-Plan2 59

17 Concluding Reinarks 61

References 62

ix

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (12)

List of Figures

1 Communication between ( ‘ c n f i ~ i l 1 ) I a i i i i ~ ~ i ~ a i r t l I*:scc.iition . Zgrnt . . . . . . . . . . 11

2 O-Plan2 Arrliitectilre . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3 Internal Structure of the ( ‘ i i i ~ i ~ c i i t O-Plan2 I’laiinci~ . . . . . . . . . . . . . . . . . 27

3 Example of activity planner at .ID\ i ih i i ig TPK . . . . . . . . . . . . . . . . . . . . 35

5 Example of resource allocation at . ins i i h i l i g T P N . . . . . . . . . . . . . . . . . . 3.5

G Example Output of tlie AutoC‘.lD-based U h c r Interface . . . . . . . . . . . . . . 33

7 Examplc Developer Interface for t h e O-I’lan2 I’laniii~ig Agent . . . . . . . . . . . 4.5

8 Commnnications \Viring l l i ~ i ~ ~ ~ ~ ~ \ \ of’ t I iv I I \ \ I h a i c l l i i c . . . . . . . . . . . . . . . 55

X

...

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (13)

1 Summary

Plaiining, schediiling and control syhteiits based on artificial intelligence techniques are now maturing and are being applied to progrehsively more realistic problems. The Knowledge-based Planning and Scheduling Group at the Artificial Jiitelligence Applications Institute at tlie Uni- versity of Edinburgh has been involvcd in the production of several complete working Artificial Intelligence ( A I ) planning systems with gradually improving scope and capability. The latest is the 0-Plan2 Architecture, the 0-Plan2 Planner based on this architecture and a demonstration environment for inter-agent coiiiiiiantl. planniiig ant1 execution using the architecture.

1.1 Project Aiiiis

The 0-Plan2 project has the followi~ig itinl:,:

0 to provide a generic domain intlt~pentlent coinputational architecture suitable for speciali- satioii into coiiimaiid, plaiining and execution q~r:,tein\ with the addition of new processing capabilities a i d doin ain knowletl g:”.

,

0 to provide a state-of-the-art .\I planning 5ystcni nhich uses an activity based plan repre- senta tion.

0 to provide iiieaiis t o allow a rich lc~.el of tloinain Iiiio~ledge to be provided to the system and to exploit this domain infoi.niatioii i ~ i opportune ways within the system when choices are heiiig made and alteriiativt.:, osploi~ctl.

0 to clarify and define the reqiiirctl 1riot l1i l0s and interfaceh of the architecture, the planner and other parts of the s;vstcni.

0 t o provide a portable and flwiblc pmlotype syhteni in which new functionality caii be experimented with. The desigii i - inteutlcd to al1oIv for experiiiientatioii with real-time distributed command. planning a n d control in a inrilti-processor computer based system in future.

0 to demonstrate tlie architect itrc iIt1tl p l a n i i c ~ r o i i realistic problems.

1.2 Project Acliieveiiieiits

0-Plan2 provides an Archifccfciru i l l ivhjch ( 1 i f f w A i i t agents with command (job assignment), planning and execution monitoring t~)lc:, caii I ) ( > 1 ) i i i l t . The architecture sr~cl<s to separate out the following components:

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (14)

i e the decision making about \vliiii the agcnt shoiiltl do nest (in the C‘ott trdl tr) , and

e tlie handling of communicatioii Iwtiveeti on(’ agent and others.

The main contrihutioii of the 0 - P l a n 2 research Iia5 been in providing a complete vision of a more modular and flexible planning and control s y h t ~ i n incorporatiiig A I methods. This report is intended to describe this main coirt I il)utioii i i i (Iviiiil.

A “state-of-project” prototype of O-l’laii2 lrrih I ) c v i i pimvitlctl which is a complete, even though simplified, demonstration of our vihioti of a iiiulii-agelii 5~ 5 t c i i i \vliere agrtit 5 are based upon tlic 0-Plan2 architectlire and where cotiiiiiiiiiicai i o t i 1)ctucwi t l i e tIiro(> agents for job ashignment, planning arid execution monitoring i \ i i r a t~~giilni f o t t t idi .

Most effort in tlie current 0-Plan2 protot! pv ha\ I ~ C C I I clt\\oted to the provi5ion of a planner which uses a hierarchical partially o r t l o ~ ~ d act ivii!. ~cl,i’c~~c~trtaiioii of plaii5 as its hasis. The aim has been to replicate tlie functionalit> o f c ’ n i liar I:tliiil)~ti~gli plCinticr5 sacli ah Nonlin [39] and 0-Plan1 [lo] but in an improvcd co i i i l ) i i i a t i o i i < i l i i .a t i i (w 0 t . k \vliicli i5 mor(’ flcsihlc ant1 caii l)c made more widely available thaii tliow (~iirl i(~t 5 ) \ t ( I t i i \ .

The prototype of 0-Plan2 inclut1c.s iiiiiiiI)(ii o f ~ ~ ~ I I I ] ) I ( ~ application doiiiain descriptions and demonstration files to show 0-Plati2 i i r iiw. .I d e t i i o t i ~ 1 i r i t ioti of t l i r . iiiteiitlctl itscr interface for 0-Plan2 has been created which 1 1 ~ 5 t I I P I\ itl(.l> ~ i \ x i l ~ l ) l ( ~ .\uta( ‘.I11 p”clidg(3 [4] to sho\v lion. the system caii link to such pack;rgc.\.

A demonstration of spacecraft plaiini 11% and cxccut ioir iiiotiitoriiig has heen created for a simple, but realistic, spacecraft model b a w l on r l l i t i i n l siiclliic.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (15)

. I

2 Introduction

Tlie research on 0-Plan2 lias its root\ i i i earlier ivorli on other Edinburgh A I planners: Nonlin and 0 -P lan l . It has drawn heavily on the experience gained over the last 20 years in AI planning research. The report begins by clra\ving together a number of important advances and individual items of technology 1% liich have beeu integrated in the 0-Plan2 design. New work on the ways i n which commantl. planning i l l i d control agents interact in a distributed, Iiierarcliical problem solving framc\\orli is de~cribctl along with tlie representation of plans as ii\ed for communication between those agcntb. 0 -P lan2 is intended to be relevant to future parallel processing platforms and for applications where the command, planning and execution agents are spatially separated ( p e r h a p i r r i t h long or irregular communication times). Hence, the new features of tlie 0-Plan2 design ititcliidcvl to allow for the management of the A I planning proct\s a4 a niimber of separate coi i ( III i ( ~ i r t ( o t i i l ) i t t < i t ioiis is tl(~scribetl.

\I.itli this background, the report t Ii(1ti ( I ~ s c I ibe\ t I I O O-l’lil1l2 architecture hy introducing the .5 iiiajor components in tlie arcliitcctii I e: I\: tion I(dgc\ S o u tw\ ant1 their coiiipittat ional Platforms; Domain Inforiiiation; the Plan Statr: t Iic ( ‘ o t i t t d l ~ r : a i i d t lie (’oiibtraint hIanagers and Support Itoiitines. These will lie referred to t Iirottglrorit the tcport dticl greater tlctail of tlie various coiriponents are the subject of Iatcit i ( v t ioti\.

‘rhc ciirrent 0-Plan2 project ha5 ('air( ( x i i t i < i t ( d oii t Iic j)to\ ision of a plaiinitig agent witliin the 0-Plan2 architecture. This i s t l i r \ i t l ) j v r i of t Iiv iiost wctioii i i \ the report. It is in this section that a description is given of the I V ~ J L, i t 1 n I i i c Ii t Iic .5 totiiponents of the architecture referred to above are spccialised to enable t l i c \ysteiii to pwfor t i t a s a plan~ier. T h e are hrief sections to de\cribe tlie simple job a4signiii(>iit ( ( o i i ~ i i i n i i t l ) r ~ ~ ~ ( l c>secution systeni agents wliicli form a part of the cnrrent 0-Plan2 protot j.1~.

‘I’lie User Interface to tlie 0-Plan;! \ j . \ tv i i l Ii<i\ 1 ) c ~ n tlcsigncd i n such a ivay that it will allow integration with a ~iiiinher of other \oplii\ticiitod I I \ C I tool\. ’Tlie nest section of the report thus Iiigliliglits the issues of user roles nit 11 i t lspc( t to ii ( ~ o t i i t i i a ~ i d . planning a n d control system and explains the may i n which 0-Plan;! ( . l i r i i act (’I is(>\ iiwi interactions. The sectioii also describes tlic interfaces built for the current O- l ’ l~ i i i2 j ~ l ~ i t i i i ~ t ~ agciit prototype.

O-PIan2 has been designed in sii( 11 I\ <IF 1 lid1 ( o i i i p o i i e n ~ s can be i~iiproved within the speci- fications adopted. Performance i \ \ i to \ liai.(> I ) o (~ i i (miisidpi c ~ l i n establishing the interfaces and protocols used. Thci cnrreiit prototypv ol tc i i i i i c Ii itl(.\ o i i l ) Y C ~ J \iiiiplc inij~ltmeiitatioiis of some o f ‘ t h ~ component\. llo\vever. estcii\i\.(> i i i \ t I II iii(’irl ,it i o i i (iiid tlidgiiostic farilities have been built into 0-Plan2 to allo\t for c s i )~ i i t~ i (~ i r i~~ i io i i i i i Itit IIIC.

l’lie main tlienie of t l i c O-I’laii2 I ( W \ < I I ( 11 Iiri\ I)(YII t Ii(1 i ( l v t i t ificatioii of wparal)le support mod- ules. internal and external intcrfac.c~ 5pw i f i c . i t i o i i \ , r i i t l i)iotoc oI\ governing procei5ing 1,ehaviourq wliic 11 are relevant to air 11 pI<iiitiiiiK \ \ \ t (iiii. l l o i i ( ~ . 1 I I P \ . o r i o i l \ coiitiiI)irtiotis ivllicli will have b c c ~ i i iiitrodiiced i n cailicr sc(tioti \ ol 1 I I V i c > i ) o r t < I I ( I ( I I ~ I H II togc.thrt.

The t i t le of the pro.j(v t - “Spdcec 1,111 ( ‘ o i t r i i i ~ i i r ~ l <III(I (‘otitrol C7siiig A I l’latining Tecliniyues” - r(.flcctcd a chosen application arcs t o (I(>iiioii\t r < i t f’ t IIP i t lws Iwing developed within 0-Pla112. The \pacecraft planniirg a n d coiit rol ( I o i l i i i i t i foi iiicd n i i s (> f i i l c.;ample wi th in which to consider t h e n r w l to separa t t~ fiitictioii~ility i n t l i f I ( ~ t ( ~ i i t <iy(>ilt\ n i t 11 I V I J tlifF(>rcllt coniplttation and real-

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (16)

time response requirements. A description is given of an O-Plan:! application to a simple, but realistic, spacecraft.

The report is orgaiiised into the following sectionq:

Section 3 relates tlie background to tlie O-Plan:! work and the teclinical influences which have been drawn upoilirn the work.

Section 4 describes our philosophy for a regular style of cominunication between agenth in a simple command, planning and control environment;

Section 5 describes the representation of a l)li>n \r-ithiii O-Plan2;

Section 6 explains the i-neclianisiiiy i t h e d in O-I~laii2 for managing concurrelit rouiputations and deciding on the order of p roce~4ng :

Section 7 describes the major components of I I I ( ~ O-Plan:! architecturc:

Section 8 goes into greater detail on how the plaiiiiiiig agent has heen provided ill tlie 0-Plan2 architecture;

Sections 9 and 10 outline the job assignment and execution systems in O-Plan2:

Section 11 describes the user interface wliicli has 1)ec.n designed for 0-Plan2:

Section 12 looks at performance issneh and tlic instriimeut at ion of the O-Plan2 prototype;

Section 13 summarises the varioiih ahpects that relate to the ~notlularity. interfaces and inter- nal protocols within O-Plan2 - an important i>hpect of the design;

t

Section 14 describes a n application of the O-Plaii2 hyyteiii to a simple. but realistic. spacecraft command and control example.

Tlie report concludes with a descriptioii of ~ c l a t c d pi'ojects and our phiis for the future.

4

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (17)

3 History and Technical Influences

0-P lan was initially conceived as a project to 1)rovide an environment for specification, gener- ation, interaction with, and executioii o f activity planx. 0-Plan is intended to be a domain- independent general planning and coiit id fl.iliii(\\t.ot.k wi th the ability to emhecl detailed knowl- edge of the domain.

0-Plan grew out of the experiences of ot her I (\\(’iIrcli into X I planning, particularly with Noillin [39] and “blackboard” systems [29]. TIi(~ I Z r t r t l / / / ~ p iu l’louni/?g volume [ 11 includes a taxonomy of earlier planning systems which places O-l’lilti i n relation to the influences 011 its design. It is assumed that the reader is familiar n.i th thew wail\-\ ah tlie bibliography does not cover all of thcni. The same volume [l] inclntles aii i t i l iwtliictioii to the literature of A I planning.

I hc main A I planning techiiiqiies \vltic.li l t ~ i \ ~ 1 ) r c ~ t i i t \ c ~ l or c~stcatled in 0- l ’ l a ~ ~ are: r ,

0 .I hierarchical planning systcttt n liicli ( ;III ~ ) t w I i i ( o I ) l i i l i \ a \ partial orclcrs on actions (as suggested I)? Sacerdoti in t l i t \o \ I I 1’1ii tiiicr [:<:J]). t Iioiigh 0-Plan i:, flexible coiiceriiiiig the ordrr i n wliich parts of t h r p l a i i i i t (Iiffer(3iit levels a l e espandetl.

0 i l n agenda-based control arcliitcc t I I I~ ’ i i i u I i i c l i c w c . 1 1 control cycle can post pendiiig tasks during plan generation. Thcw p(>iidiiig t a \k \ i\r(’ tlieii picked np from the agenda and processed I)? appropriate h a n t l l o ~ ~ ( 11 i ; . \ i < \ . \ j - 1 1 [;! I] aiit l O P N [‘L’2] iih(’5 the term Tinowledge ,S‘ourcc for these handlers).

0 The notion of a “plan state“ \v!iicIi i \ t I i ( 3 (liltii \trii(titre containing the emerging plan, the “flaws“ reinaiiiiiig in i t , i l l it1 t l i ( > i t i l o r t i i a t io i t I I \ ( Y I i n I)oilding the plan. This is siiiiilar to the work of hlrDermott [’LS].

3.1 Q-Plan1

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (18)

methods, improved heuristics and tctlriiiqiic~s for swrcli space control. and a deiiioiistratioii system embodying the results in a n appropriatc fraiiie\vorli a n d representational scheme.

0-Plan1 began with the objective of hniltliiig aii opein architecture for aii A I plaiiiiiiig project with the objective of incrementally developing a system resilient to change. It was our aim at the start of the project to build a \\..stem iu uhich it was possible to experiment with and integrate developing ideas. Further. the tem was to be able to be tailored to suit particular applications.

t

3.2 0-Plan2

The 0-Pla.112 project began in 1989 and had tlie following new objectives:

to consider a simple “three agent” view of the enviroiinient for the rehearch to clarify thinking on the roles of the i i w r ( s ) . architecture and system. The three agents are the job assignment agent, the plaiiiiiiig agont ai i t l t h e execution agent.

to explore the thesis t h at coin 1 i i uni c a tion of ca pa hi li t ies a ii d i iiforiiiat ioii bet ween the three agents could be in the forin of ylor, p n f c l ~ c . c \vhicb in their turn are i n the same forin as the domain information descriptions. th(. task dcscription and tlie plan representation used within the planner aiid the> other i n o ageiith.

to investigate a single architecture that coiild siipport all three agent types and which could support different plan i.epicsciitatioii:, aut1 i t g ~ t ~ t rapa1,ility tlcscriptioiih to allow for work in task plaiiniiig or r e s o i i ~ ~ c sclictluling.

to clarify the functions of coiiiporlonts 0 1 i t plaiiiiing anti control architecture.

to draw on the 0-Plan1 esperit.iitc aiitl to iuipi o \ ~ 0 1 1 it especially wi th respect t o flow of control [ l a ] .

to provide an improved version of t lie 0-l’ lni i s j st(wi suitable for iise outside of Edinburgh within Coininon Lisp, X-i\’iiidon n n t l I \ I \ .

to provide a design suited to i t s o on pai.iillel proec4ng systems in future.

The first 0 -P lan project at Ecli-Hburgli. 19s I- 19SS. f a a s ~ c c l on tlie techniques and technologies necessary to support the informcd \earcli pro( c s needed to generate predictive plans for siibsequent execution by some agcut . ‘ r h e O-l’laii2 pi~)jcct continues the emphasis placed on the design of a planning and control arcliitcctii re itlciitifyiiig tlie niodular functionality, the roles of these modules, aiid their software in t t~ i~ facc~ \ , O-I’laii2 hi>\ resulted in a denionstrator, capable of acting as a foundation for furthcr tlt~i~olopiiicnt. iu atltli tion to descriptions of the underlying sub-systems and modules which U P fwI i i i c i t i i p o i t i i n t l o siil)port il practical planner.

0-Plan2 is incorporated n.ithiii a l ) l < t ~ I i l ) ~ < > rcl-li Le frniii(\nor.Ii: for efricioiicy reasons ive have chosen an agenda driven arc1~itectiii.c. I t r i l l \ 011 t Iic dpciidas i~~preseu t outstancling tasks to be performed during the planning prot~1>s. l ~ ~ t ( l i l ivj i v I ; r t ( > c l i i t ~ t l y to the set of f l ( ( i ~ . < identified as existing withiii the emerging plan. .\ \i t i i l ) l c > v ~ ~ i t t i l ) I o of i t j7~cr. is I h a t of a condition anaiting

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (19)

satisfaction, or an action requiring refincriiciit to planning cycle which flaw to operat(> 011 i icxt .

The nature of these flaw types has been infliienccd ti\. experience from tlie 0-Plan1 work, but the iliain development focus is the handling and processing of the flaws. The "knowledge sources" employed in @Plan2 have cleaner triggering mecha~iisms and have been given a variable level of granularity, enabling processing to b e siihpendcd i f needed (we refer to this as knowledge source staging) while further flaw information i5 gathered. This is particularly useful for a planning system which attempts to he opport nnihtic and to operate on a least coininitinelit liasis, while retaining completeness of search (~vliere possible). It will also siiiiplify the task of maintaining and reasoning with partially bound variables in the plan, which proved to be dificult and limiting in tlie 0-Plan I n.orl;.

Research in 0-Plan2 has been coiic('ii1 I atiiig o i i 1110 problcnrs associated with:

lonrer level. A controller chooses on each

0 temporal constraints and rea5oiiiiig. ' I ' l i ( ~ uiitlerl~~ing data structures have been con- pletely re-designed and reworkctl fro111 t I i v 0 - P l a n I l\rorli to allon. €iirtlier development of the temporal search based pi l i l t iiig aIgoi.ithin\. aiid to h~ppor t the enhanced condition achievement procedure.

0 resource utilisation inanagentelit . l < c \ o i i r co\ pro\,itle the most olivious l ink to scheduling, where successes in resoiirce i i t i l i \ a t io11 ~ii~iiiiigeiii(~~it lia\.c heeii niore pronounced. though still limited.

0 plan control. 0-Plan2 is iiitc~irtlctl to (oiitiiiiiiiic.ritc pIaii5 to an esecntion agent ivlio can coiiiinunicate progress back. ('o;itroI ht 1 at(>gip\ are therefore required to enable plans to be repaired in the case of simple. failiirc> or to hcgin replanning if required. Earlier work employing qualitative proce\b [ 1 .$] t Ii(wi \ \ \ i l l ; l\\i\t \ \ r i t l i repair strategies in future.

The end goal is t o be able to deiiionstralc> a t l O l J l i l i l l iiidependent --!I Planner capable of accepting descriptions of planning doniains and gcitcral iiig 1 ~ a I i 5 t i c plans for suhseqnent execution.

3.3 Characterisatioii of 0-Plan2

The 0-Plan2 approach to coniiiiaiitl. 1)liil~iiilig. \( lidiiliiig aiitl control can he cltaracterised as folloTvh:

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (20)

- time point network maiiagc'r

- object/va.ria.ble manager

0 using localised search to explorc altcriiati\.w \vliore advisahle

0 with global alternative re-orient at ion \vli~re necessary.

0-Plan2 is aimed t o be relevant to the fol10\riiig types of problems:

0 project management for prod u c t in t rodu c t i 011, sys t enis eiigi n eeri ng , cons t rii ct ion, process flow for assembly, integration arid verification. ctc.

0 planning and control of supply and t l i h l r i h u l ioti Iogihfics.

0 iiiissioii sequencing and contid of h j ) i i ( ' ( ' 1 ) i ~ ) I ) ~ h h i i c l i ah \'oj.ager. S R S - I . etc.

These applica tioiis fit mid way b etlvee ii t lie 1 il rgc) h c i ~ l ~ ~ 111 an u fac t uri ng .sc.licduling problems found in some industries (where there arc oft(1n fc\v iiit(~-opcrat ion coris t~~aints) and the complex puzzles dealt with by very flexible logic 1)ahetl t o o l h . I lowver. the problcms of this type represent an imp or t ant class of in dust ri a1 re1 cva 11 cc .

S

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (21)

4 Communication in Command, Planning and Control

The aim of this section is t o clescribc i n broad terms the motivation and reasoning behind the design of the 0-Plan2 architecture. Edinburgh research on planning and control architectures is aimed a t building a practical prototype system which can generate plans and can reliably execute the plans in the face of simple plan failures.

We are using our experiences in dealing with applications of AI planning techniques to practical projects t o develop a planning system that closes tlie loop between planning and executing. There have been some successes with previous attempts at closing the loop [13], [18], [27], [4], but often the plans generated were rather liinited and not very flexible. In general, the complexities of the individual tasks of plan representation, generation, execution monitoring and repair has led to research into each of these issues separately. In particular, there is now a niisinatcli between the scale and capaljilities of plan representations proposed for real-time execution systems [20], 1301 [32], and thaw that call be generated by today’s A I planners. However, in most realistic domains the dei11a1id i:, for a system that can take a coininand request, generate a plan, execute it and react to simple failures of that plan, either by repairing it or by re-planning. Explicit knoivledge about the structure of tlie plan, the contribution of the actions involved and the reasons for performing plan modifications a t various stages of the plan construction process, provides us with much of tlie information required for dealing with plan failures. Such knowledge is also ebsw tial for furl her planning and re-planning by identifying generalisations or contingencies that caii tic i i i t i ~ o t l ~ i c c ~ l into the plan in order to avoid similar failurcs.

One of the largest simplifications most plaiiiiers to (late have made is t o assume plans are constructed with full knowledge of tlic capahilit ie:, of tlie devices under their control. Thus, executing such plans involves the direct application of tlie activities within the plan by an execution agent which has no planning capahilitj.. TTnfortunately, unforeseen events will occur causing failure of the current plan and a wqii(+t for repair of the plan or re-planning directed at the planning system. Building into the execution agent some ability to repair plans and to perform re-planning would improve the problcin solving performance of the execution agent, especially when it is remote froiu the c w t ral 1)laiiniiig systein.

4.1 The Sceiiario

The scenario we a,re investiga,ting is a.s fol1on.s:

0 A user specifies a task that is to ljc pci.forined through some suitahle interface. We call this process job assig12172 t nt.

0 ,4 ylccmer plans and (if requehted) arranges to execute the plan to perform the task specified. The planiier has knov lrtlgc ol‘ tlie general capabilities of a semi-autonomous execution system but does not i i c c t l to kiion aljorit t h e actual actil’itie5 that execute the actions required to carry out t l i v doti iwl t ri51,.

0 The erecufiou tcm seeks to car1.j oiii i IIP tlctailecl t ]<s specifietl 1)y the planiier while working with a more detailrtl i i i o ( I ( ~ l or‘ t I i ( 1 c i ~ t c n t i o i i tiivironitient t h a n is availahle to the

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (22)

job assigner and to the plaiincr

We have deliberately simplified our coiisitlc~r,ii i o i i i o i Iri~cc agents wi th i l i c w different roles and with possible differences of requireniciiih for i i \oi ii\.iiilal)ility. processing capacity and real-time reaction to clarify the research ohjectivo\ i t i o i i r \ i o r k .

The executioii agent executes the plan hy c.lioo\ing the appropriate activities to achieve the various sub-tasks within tlie plan, u\iiig i t \ 1;iirnvlcdgc about t l ic particular resources under its control. Thus, the central planner coiiiiiiuiiic;?t(~h a general plan to achieve a particular task, aiid responds to failures fed hack from the esc~cution agent which are iii the form of flaws i n the plan. The executioii agent commuiiicateh Iv i th t 1ic real world by executing the activities within the plan and responding to failures fed back froni tlie real world. Siicli failures may be due to the iiiappropriatenes~ of a particular acti\.it).. or I)pcii1l\c the desired effect of an activity was not achieved due to an unforeseen c\.ciii. ‘l’lic~ I (’;?yoti for i I i r failure dictat(1s whether tlie same activity should he re-applied. rep1,iccd \\ i t 11 oi Iict. r l ( i i\.itiw or \vlict 1 i ( ~ ro-planiiing slioiiltl take place.

i

#

4.2 Use of Dependencies

4.3 A C o 111 111 on Repr es e 11 tat i o 11 fo 1- C o in111 11 11 i c a t io 11 b et w ee 11 A ge 11 t s

I O

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (23)

case. the communication l)et\vecti t I i c cotit t.ii1 plaiiii(tr atid the execiit ioii agent becomes an i t i t cresting research issue.

I P1a.n State

/ , I lie (oiiiiiioii representation include\ liiio\vlccIgo about t lie capabilities of‘ the planner aiid ese- clition agent, the reclitirementx of i l i r pliiii r i t i d t Ii(. plaii i t \elf‘ either with or \vithout flaws (see l’igure 1). Thus. a planner will respoi~(l to t IIC t q i i i r ~ ~ i i i e i i ~ ~ of a i i s ~ r . Babetl oil the knowledge of i t 5 ow11 capabilities and that of i l i ~ (>xociit ioii (’II\ iimiiirnt. it will generate a plan. This plan may then be executed directly i n t l i c l’(’iil 1\01 I t l . o r . iiidiiwtly via an cserutioii agent. The csccittion agent executes this plaii i i i t Iio i ~ ~ n l vi odd i i l i t l inonitorb the execution, responding to failures in one of two ways. If i t doc15 iioi li<i\(> kiiowlcdge of its own capahilities, it simply rctiirns knowledge of the failure t o t l i v c c t i t i*dI l ) l ~ i i i i i ( ~ i ~ and awaits a r e v i d plan to be sent. In thi5 cahe, the execution agent i5 (Iiiiiil). I I i t (lo(’\ have liiioivledge of i t \ owii capabilities, i t m a y attempt to repair the plati i i ~ l ( l 111oi1 (otifitirrr \ \ i t 11 csecatiou. On the other hand, if ii repair ih beyond the capabilitic.5 of i l i t . P Y ( Y I i f i o l i ageiit. t l i en this 1inon.Ic~Ige is fed h a c k to t h o central planner a n d again a r(xviwtl I)lciii i5 c ’ u p c ~ t r d . I n this case. t lie execii tion agent is hcnii-iiiitonomoiib. Wlirn failiii.e> t l i i i iiig i I I V opl) l i ( ~ i i i o t i of‘ t Iic plan arc’ f c d h a c k to the planner, tlicw tiiay lie acted npon by i t a n d ii t t I p < i i i of t l i t i p l ~ r i iiiatle or total re-planning instigated. This may, in turn, involve the iiser i t i ic\f‘or tiii i ldt itig ilie tahli requiretneiit. -4 revised or new 1) I i i t i i5 then excciitetl. Finally, ~ I I ( ’ ( P ~ ~ of t l i ( ’ ( i \ ( v iitioii o r piii~ial csccution of the plan is fed b a c k to the iiser.

0t l ic . r issues rcliiiing to the choice. of t I i ( 1 ( o i i i i ~ i ~ ~ i i t (>i ) iwoi i t : i t ioii and comtiiitnication protocols i II c l ii (1 e:

I 1

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (24)

5 Representing and Communicating Plans

'5.1 Plan States

One of the most important problems which needs to be addressed in any planning system is that of plan representation. An 0-Plan;! agent's pin/, stcrte holds a coiiiplete description of a plan at some level of abstraction. The plan state also contains a list of the current f k ~ w s in the plan. Such flaws could relate t o abstract actions that still must be expanded before tlie plan is considered valid for passing on for execution. unsatisfied conditions. unresolved interactions, overcommitments of resource, tiine constraint faults. etc. The Plan State can thus stand alone from the control structure of the A1 planner in that it can be saved and restored, passed to another agent, etc.

At any stage, a plan state represents an a h t r a c t view of a set of actual plans that could be generated witliin the constraints it contaiiib. -Alternative lower level actions, alternative actioii orderiiigs and object selections, and so on i1re aggregated within a high level Plan State descriptioii.

5.1.1 Task F o r m a l i s m (TF)

Tri.4 Formalism (TF) (as used in Nonliii ant1 0-f'lanl) is a declarative language for expressing action schemata, for describing task requebt5 and for representing the final plan. It allows tiine and resource constraints in the domain t o lic iriotlelled. The planner can take a plan state as a requirement (created by a T F C'oiiipiler from I lie 1 1 ~ 1 ' providd task specification in T F ) and caii use a library of action schemata or generic plan \late fragnients (themselves created Iiy tlie TF C'ompiler from a domaiii description provided 11). the user) to transform the initial plan state into one considered suitable for teniiinatioii. l'liis fiiial plan biate could it\elf be decoiiipiled back into a TF description if required.

Our design intention for 0-Plan2 is that a i i~ . 1) l i l l i 5tatc (not jubt the initial task) can be created from a TF description and vice verba. Titi, \ \ < I \ i iot 1~11ly achieved in 111(. 0-Plan1 prototype [lo], but this remains our goal.

The 0-Plan2 design allows for differelit plil11 51 a l e reprcwiitations in the different agents. Task Formalism is particularly suited to t h c rcyrtwiitat ioii of a plaii state within the planner agent and, hence, t o act ns a basis for coniniiinic;itioIt to the plaiiiier's superior (jot) assignment) and subordinate (execution system) agent\. The actual plaii state inside the job assignment and execution system agents is likely to diffcr IO that within tlie planner. For example, the execution system may be based on iiiorc pro( d i i rill representations as are found i i i languages like PRS ( the Procedural Reasoning SJ e111 [ 'LO]) , i i i d ilia) alloir. itei*atioii. contlitioiialb, etc.

5.1.2 P l a n F laws

The plan state cannot contain arhit rarj (lCit ;I o l ~ ~ i i r t ~ i i ~ ~ . 'rlie . \ I pla1iiic.r is iitade up of code that caii interpret tlie plan state ( 1 ~ 1 ; i \ t t i 1 ( 1 i i i ( ~ < t r i t l i r r t o t p i c t thc lists o f I1an.s in such a way

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (25)

that i t caii select from amongst its conipiitatioiial capabilities ancl its library of domain specific information to seek to transform the current Plan State i t is given into something that is desired h y the overall architecture. This is defined as the reduction of the list of f7uw.q known to the planner. The 0-Plan2 architecture associateh a Iinowledge Source with each flaw type that can hc processed [9]. An agenda of outstanding flaws is maintained in a Plan State and appropriate Knowledge Sources are scheduled on the basis of this.

In practice, the O-Plan2 architecture is deigned for operation in an environment where the ultimate aim of termination will not he achieved. There will be new coiniiiand requests arriving a n d earlier ones being modified, parts of plans will he under execution as other parts are being elaborated, execution faults are being Itandled. etc.

Lie believe that the basic notions descrihed above can serve us well as a basis for ail attack on the problem of coordi na tcd coiiiiiiaiicl . p I i f iiii i ug ii ntl exec 11 t ion in con tinuouhl y ope rat i i I g doniai 11s.

There miist be a means iiicremeiitally to coitimuiiicate plan related informat ioii hetween the agents involved with commanding. plaiining and executing plans - each of n I i i c 11 \ \ i l l have their o ~ v n level of model of the current coiiiiiiaiid en\ ironnient. plan and execut ion ciiviim~iiiieiit. We will explore the properties that we niust >eel< from our hasic notions in the following sections.

5.2 Plan Patches

The requirement for asynchronouhiy oilcrating planner5 and execution agents ( H tit1 i iidcwl users and tlie real world) means that it i5 iiot appropriate to consider that a plau iwluirenieiit is set, passed on for elahoration to the plaiiner aiid then conimunicated t o a wait ing which will seek to perform the act ions involved. Instead, all components must he considered to be operating independently and iiiaiii tainiiig themselves in sonie stable mode \r.hcre they are responsive to requests for action fiotii thc othei. coiiiponeiits. For example. the execiitjon agent i i~ay have quite elaborate local mcclianimi5 and inhtructions to enable it to maintain a device (hay a spacecraft or a manufacturing cell) in a safe. healthy, responsive b t a t r . Thc task then is t o coiiiniunicate some change that i5 rcquebtetl from oiie component to anot Iicr aiid to insert an appropriate alteration in the recoi\cr such tha t the tasks required are ciirl.i(It1 ou t .

lye define a Pkln PNtrh as a liiodificc~ ler+xi of the type of P h i State i i w I i t i O-I’lanl. It liah sonie sjiiiilarity to an operator or action expansion bchema giveii to ai, . \ I pla~iiijiig system i n that it i:, an abstracted or high I c \ ( ~ i ep i t~cn t~ i t iou of a part of the t a sk t l i a t i \ required of t hc receiver using tcrininology releva t i t to tlie I cwiver‘5 capabilities. Thi5 provides a simplified or I)lack-hos view of possilily qu i t c dot c i i l c d i i i 5 t rite ti0115 needed to act ually p~ i~ fo r i i i tlie action (possihly involving itrrators and conditional>. c t c ) . (’oiiiplex execution agent rrpresentatioiial and programming languages can lw l i c i t i t l l c d 1)) ii5iiig t hi> abstracted viclr. (e.g.. [20], [30] ). For ex a 111 111 r , re1 i ahlc t a sli a ch i rvi II g br h (I r i o I I t.< Ii i c I i i i i c I 11 cl e (1 con t in gen ci c s i1 ti ( I sa fe stat e pat lis to t l t d xvith unforeseen events coiiltl i )v fiitltlcii from the. plaiiiier b!. com~iiuiiic <i t i o i i i n ternis of a himplificd and more robust niotlcl ol t I i c cxcv.rit ion operations [ 2 T ] .

Outhtanding flaws i n the Plau P a t ( Ii < I I ( ’ c~oni~ii~iiiicatctl along ni th the 1)at cli iiwlf. However, these flaws must lit those that cat i I)(’ Ii i i i i t l lrt l 1)). the recc’ vver.

I t caii 1)t seen tha t the arraiigeiwtit ~ l ) o \ ~ ( i i i o b t I! a35nnicd to refer to t h o ~oiii~iiiiiiicatioii be-

l :{

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (26)

tween a planner aiid execution agcwt ) albo reflect\ the communication tliat takes I)liice between a user and the planner in an 0-PIiiti’L type . \ I planner. Requiring rattier inore effort is the investigation of suitable Plan Patch cotihtryct\ to allow esecutioii error:, to be passed hack to tlie planner or information to be p a h w d back to tlie user, but we believe t l ia t this is a realistic objective.

5.3 Plan Patch Attacliiiieiit Points

There i s a need t o communicate thc pointh at which the Plan Patch shoiiltl 1)e at tached into the full Plan State in the receiver. ’I’lie sender and receiver will be operating asynchronously and one side must not make iiiireamiitble assuiiiptions about the internal s ta te of t lie other.

We endow all the components n-jth rl iwil-linio clock that can be assiimecl to hc ful ly 5yiicliro- nised. We also make simplifying c t s h i i i i i p t i o i ~ h allout dela in conimiinicat i o i i to k c c p to the immediate problem we are seeking to t aclile (n-hile fiilly believing tliat cstciisioii to (.iiviron- nients where communication delay i > involved will be possible). Ther~f‘ore. i i r r t ric t i i i i c is the b‘l)ack-stop” as a means of attacliiirg ii Plan Patch into the internal Plan Slat(. of‘ t l i ~ wwiver. Metric time is also important to start 1 Iring\ off a n d to ensure a coiiiiiioti t ~ ~ f ~ r ( ~ i i ( . ~ I’oiiit when necessary (e.g., in cases of loss of m i i t 1.01 ).

However, the use of metric time a5 [in nttaclinient point lacks flesibilit!.. I t give\ t hc receiver little information about tlie real iiitcvit io115 behind the orderiiigs placed on t IIC cwiiipoii(’iits of the Plan Patch. It will, in some c a w \ . l)c better to conimunicate in a relati1.c or cI(titlific>tl way to give the receiver more flexibility. Siiitahlc forin.; of flexible Plan Patch .\I tilcltt~i(3lit Point description will be investigated iii f l i t I I I C ( \uc l i ah debtriptions relative to t Ii(1 o s p c c t c ~ l Goal Stnicture [39] of the receiver).

5.4 Iiicreiiieiital Plan States

0 a plan patch.

0 plan patch flaws as an agetitlci of ~)e~i~Liiig t a 4 i h .

0 plan patch attachment pointh.

Such Incremental Plan States arc i i w l tor t n o \i . comiiinnicatioii 1)etn.coii 1 I I C i i ~ r and the planner and between the planner ;itit1 t l i t escc ntioii agent. The 0 - P l a n 2 T’liiii S t a t c st nictures and flaw repertoire has been e s t w ( l ( 4 to c~)pc. initiall!.. vi it11 a dunil, ( > x ( ~ i i t i o i r <igviit t l i i i t can simply dispatch actions to he carricid o i i t <ltlitl i.cc.c>ii.c fault reports agnirt5t ii i i o i i i i i l i i t ( d b e t of conditions to be explicitly monitoictl ( < I \ tlc>cril)cd i i i [ 101). In future i ~ w a i ~ 11. t l r c l’laii State data structures aiid flaw repertoil(’ \\ i l l he oxtentled again to cope nit I r ii :,ciiri-iilitotiomons esecution agent with some capabiIit> to hi tlic>i elahorate the Incr(1tnoiital 1’1;-1ii Str i l (x \ iilicl to deal locally with re-planning req i i i i v i i i v i i t \ [:J I ] .

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (27)

A means to compile an Incremental Plan State from a modified type of Tasli Foriiialisiii (TF) declarative description (and vice versa) will be retained.

5.5 Plan Transactions

The overall architecture must ensure that an Iiicreinental Plan State can be understood by the receiver and is accepted by it for processing. This iiieaiis that all the follo~ving are understood by the receiver:

0 plan patch description is clear.

0 plan patch flaws can he liaiitlltd 1)). tlie receiver’s Knowledge Sources.

0 plaii patch attachment point\ are understood.

It is iinportant that the sender and receiver (n-hether they are the user and the A I planner, the planner and the execution agent, or one of the reverse paths) can coordinate to send and accept a proposed Incremental Plan State which the receiver must assimilate into its owii Plan State. We propose to use trnizsaction pro UQ iiietliods to eiibure that such coordiiiatioii is achieved.

We have created some specific flaw tvpes aiid Ihowledge Sources in the various components (job assignment, A I planner and execution agent) to handle the extraction aiid dispatch (as an Incremental Plan State) of a part of an internal Plan State in one coiiiponent, aiid the editing of such a n Incremental Plan State into the iiiteriial Plan State of the receiver. The “extrac- tion” Knowledge Sources inuse be supplied with iiiforiiiatioii on the Plan Patch description, flaw types and attachment points that the receiver will accept. This constitutes the priiiiary source of information about the ca1)al)ilitieh of tlie receiver that tlie sender has available and its representation will be an important part of the research.

C‘oinmunication “guards” will ensure tha t the CI priori criteria for acceptance of an Incremental Plan State for processing by the receiver’s I<non.ledge Sources are checked as part of the Plan Transaction. It may also be tlie case that initial information about urgency \Till be able to be deduced from this acceptance check t o prioritibe the ordering of the new f l a w with respect to tlie existing entries on the agenda in the receiI.er.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (28)

6 Managing Concurrent Computations

The 0-Plan2 architecture has beeir designed to allow for concurrent processing where possible. Tlie systems implementation itself is compowd of a number of parts representing the major components of the architecture. These can be run as separate processe if desired. In addition, tlie basic flow of processing performcd by the architecture allows for a uiazTcfroiit of concurrent threads of computation to be maintained and decihions can be taken about where t o deploy any computational effort available (whctlicr this is actually impleiiiented with parallel processors or

;”

not).

0-Plan1 made a start on ~neclianisnis for the iiiil)leiiientatioii of an efficient planning system able to take an opportunistic approach to selecting where computational effort should be con- centrated during planning. However. some l imi t atioiis were observered antl t alien into account during the design of 0-Plan2. The O-l’Ian2 ~ ~ ~ ( ~ c l i a i i i ~ i n s are listed in the folloning scctionh.

6.1 Choice Ordering Mechanisms in 0-Plan1

6.1.1 Building up Information in an Agenda Record

0-Plan1 included the ability to allow a linon ledge sotme to examine a possibl~ tlrcision point (represented by the agenda entry i t i 5 aslied to process) and to add information relating to tlie choice to the fields of the agenda recoid. If tlie choice did not become siiitahly tightly restricted as a result of the addition of thi5 information. i t was pohsible to put tlic agenda entry back oilto tlie outstanding flaws list with impro\cd information for deciding on the time to reselrct it for processing. Tlie ability to build 111) iiiforinatioii around a n agenda entry in an iucrenicnt a1 way prior t o a final knowledge source activation is an iniportaiit feature that e n b u i ~ s t Irat work done in accessing data bases and cliecking condition5 can be $aired as far as possible \vlicn processing is halted. There are some similariti(.s to iiiecliaiii~ms within real-time respon5ive architc~tures such as RT-1 [38].

6.1.2 Granularity of Knowledge Sources

Each knowledge source wjthin tlic 0 - P l a n architecture encodes a piece of planiiiiig kno\cledge. For example, how to expand an action. bind a 1 arial~le, cliecli a resource. etc. Fro111 a nlodularity viewpoint, there is some advantage i l l having a i.ery fine grain of knowledge so i i i~c to implenient planning knowledge. However, this caii lead to tens of agenda entrics antl 1,trov ledge source activations with the overheads associated with 5uch activations for even the simplest types of action expansion. In simpler planners. hiicli a5 Noiiliii. an expansion is efficiently handled as an atomic operation. There is a toiiflictiiig desire t o have efficient large grain I<nowlrdge sources implementing planning line\\ lctlge arid \*el J. fine grain knowledge sou I individual step of some higher lcvcl pIdn modification operator.

111 0-P lan l , with a finer grain of lilion Icdgc sourcc~. i t 1 ~ ~ 1 s also found that oidcring relationsliips between agenda entries left in t Ii(. wgviitla l i5 t hat1 to bc stated to en suit^ offiri(vit piocessing. The controller was then requirctl 10 I I trrai cl t l i t . u c>l) of actiT-at ion ortltr irig5 tha t resulted.

r

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (29)

3

A special form of agenda entry callctl a seqirciz(~ was implemented in 0-Plan1 to assist tlie controller in this task, it would only consider the head of the sequence for activation at any time, subsequently releasing tlie follon.ing agenda items clustered in the sequence in the order indicated. This process is similar t o the control blocks used in the Tecknowledge S.I system [XI ] .

6.1.3 Priority of Processing Agenda Entries

0-Plan1 assigned priorities to every flaw as it was placed on the agendas. The priorities were calculated from the flaw type, the degree of determinacy of tlie flaw and iiiforrnation built up in the Agenda Record as described earlier. These provide measures of choice within the flaw. Two heuristic measures were maintained in each agenda entry. One called Brrrnclz- 1 indicated tlie iriiinediate branching ratio for tlic dioice point. An upper bound on this can be maintained quite straightforwardly. The second ineahure n as called Brtinclz-n and gave a heuristic estimate of the number of distinct alternatives that could be generated by a iiaive and unconstrained generation of all tlie choices represeii tcd by the choice point.

In 0 -P lan l , three agendas were niaintained to efficiently select between agenda entries which were ready for knowledge source activation and ones awaiting further information t o bind open variables in the agenda information. This is described in [9]. Eventually though, the ready to ~1111 agenda entries are simply rated according to a nuiiierical priority maintained for each agenda entry on the basic, of flaw type and estimatorh which said how many choices there could be down a particular search braiich (the Brtrnch-1 aiid BrCrnclt-iz estimators). This forms too simplistic a measure for allowing tlie controller to decide between waiting agenda entries. Consideration was given t o a rule based controllcr n-itli knowledge of other meciszrrcs of opliortuni.im but no implementation of this was done within the original 0-Plan1 system.

6.2 Choice Ordering Mechanisms in 0-Plan2

0-Plan2 seeks t o provide a more colicrent set of iiiecliaiiisiiis t o enable the planning aiid control system builder to select suitable implementation inethods for describing choices, posting coil- straiiits which will restrict choice. pohtpoiiiiig choice iiialiiiig decisions until the most opportune time to make them, and triggering choices that are ready to be acted upon. These mechanisms are:

e the use of stages in knowledge hoiirceh to allow for a linear thread of computation to be defined which can be assumctl t o rim through to completion, but piovides a meaiis for interrupt ion at defined st aging points.

0 the definition of triggers on linan Ledge wurce:, and knon-ledge source stage5 to provide higher lele1 of li1~011ledge source activation checks to the a clear means to delegate

controller.

0 tlic use of compa7rnd ngf 1 7 f h f oti.)C.\ I O p u ~ direct dependencies of soiiie taslih oii others that must complete earlier. Thi5 a l l m s couiplcs coiiiputational clependencies and strategies to be created.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (30)

0 tlie use of ngenda mnnnger pr;orifies t o allow tlie controller t o select appropriate ready- to-run agenda entries and 1iiatc.h these to n.aiting knowledge soiircci platfornic;.

The following sections explain each of these niechanisms in more detail.

6.2.1 Knowledge Source Stages

The 0-Plan1 mechanism for building u p inforlnatioll in an agenda entry prior to making some selection between alternatives was a very useful feature but proved difficult to use in practice. A knowledge source had to be activated to initiate processing which might simply add a little information to the agenda entries and then suspend to allow the controller to decide whether t o progress. This is very inefficient.

In 0-Plan2, knowledge sources are defined in a series of sfriges. There can be one or I I I O ~ C stages, only latter stages may make alteration:, t o the plan state (thus locking out other knowledge source final stages which can write to the same portion of the plan state). .Any earlier stages may build up information useful to later stages. -At the end of any stage, the li1lo\vledge source must be prepared to halt processing if a\lied to by the controller. If it is aslied to halt at a stage boundary, the knowledge source may snininarise the results of its computation in a field of the agenda record provided for this piirpo:,~. -1 controller directed support routine is called hy the knowledge source at tlie end of each st ijge t o identify ivhether it must halt or may continue. This allows the controller t o dynamically rcl-direct coinputation as it considers all t lie information available to it, while providing a siinplc and efficient way for tlie knowledge source to continue coinpu t at ion without in t errnedi a t e \ t a t e aving uhile it continues to receivr a go-alrtad from tlie end of stage continuation autliorixation routine.

A Knowledge Source Fornzali.sn2 for 0-Plan2 is being designed to allow for stage definition and to assist with declaring the re\triction:, on the plan state portions affected by the final plan state modifying stage of the knon ledge \oiirce - to assist in lock iiiatiagciircn t in parallel implement at ions.

6.2.2 Knowledge Source Triggers

In 0-Plan2, a meclmnism of setting triggers on a.genda. entries for a.ctiva.t,iiig knowledge sources (aad an individual stage of a. knowletlge source if desired) is provided. The triggers limy use va.rious “items” of da.ta available within the plan st,a.te a.nd other global inforiiia.tion ava.ilable t o the planner. These may include tliings such a.s the a,vailability of a. specific binding for a. plan variable, the sa.tisfa,ction of a contlition at a. specific a.ction node in t,he plan network, the use of a specific resource, the occurrelice of an est,ernal event, inforiiia.tion froni the ‘.clock” within the planner, etc. The I\non-leclge Source Formalism referred to ea.rlier ndl also he used to define triggers 011 knowledge soli r w stages. The t,riggering constructs in the language a.re initially quite restrictive to ensiiro t efficient. a.gentIa eiit,ry triggering iiiec~~ianisms can ])e implemented. However, a,s we gain (xxpprieiicp. ive expect. t,he tiiggering langtiagc to be quit,e comprehensive. A knowledge soiirct’ 111ay also d ~ ~ ~ i a i i ~ ~ ( ~ a l l y create a. trigger oti ii continuation a.genda. entry when halting processiiig a1 a

IS

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (31)

Only agenda entries which are currently triggered will be available to the controller for decisions on which entries t o activate through to a 1;iiowledge source runiiiiig 011 a knoivledge source platform.

6.2.3 Compound Agenda Entries

Individual simple agenda entries can be grouped together into compound agenda entries. Only the head entries in the compound agenda entry are considered a t any time 11y the controller (and possibly by tlie triggering mechanism considered above), thus cutting down on the ainouiit of processing required by the controller to select tlie next agenda entry to execute when such pre-defined orderings can be specified. C'onipouiid agenda entries can be made by knowledge sources to act as a meta-processing level to iiiiplement some definite planning strategy or to implement planning algoritliiiis witli fi iier grain linowledge bources to provitlr ti1otllili~rity and real time response iinproveiiieiit .

A Support Routine is provided i n O-I'laii2 to allow any knowledge source to easily aiid reliably biiiltl and return a coiiipouiicl ageiitla entry.

6.2.4 Controller Priorities

The controller is given tlie task of tlwidiiig nhicli of tlie current set of triggered agenda eii- tries should be run on an availahlc 1;nowledge source platform. It does this by coiisideriiig tlie priority and measures of opportiini\iii of tlie agenda entry. Four priority levels are avail- able within 0-Plan2 - Low, hiletliriiii. TIigli and Emergency. The Emergency priority level is only available to handle incoining estcrnal eI.ents. The RT-1 system has siiiiilar priority based processing arrangements [38]. In ccrt ai ti case\. an O-Plan2 impleinentation \rill possess knowl- edge source platforms dedicated to processing 5pecific real-time responsive events appearing as agenda entries - thus allowing for relial)l(~ real-time rebpoiise to events categorisctl iis Emergency priority.

waiting knowledge source platform \i ill be ahle to run one. several, or all knowleclge soiirces. -1ny restriction on a specific platfoiw \vi11 he lmon-ii t o tlie controller. Only triggered agenda entries at tlie highest priority lcvcl \\ liicli can be processed on a waiting kno\vledge source are considered by the controller on eacli c ~ ~ c l e . \\*here there is still choice, a range of t n e ~ ~ u r e s of o p p o r t ~ i n i ~ ~ ~ z ciizrl priority are c~inp1o~c~I to tilalie a selection. The underlying principle is to make a selection according to a st1ati.g) gi\ ( ' t i to the controller. Initially t l i i h 5t rategy will use user selected preferences or by clcfciiiIt \ \ i l l \wl, to reduce search to the extent i t can judge this (reflecting tlie opportiiiiistic geiierdl i \ (1 pliiiining iiaturtl of tlie early xrersioiis of' 0-Plan2 - like its predecessor 0-Plaii l) . hicasurcs sitcli ah Uintrch-1 (tlie iinniediate branching ratio for the choice point) and Brcrtzclz-t? ( a I t c > i i r i \ t it. wtitnate of the number of distinct alternatives that coiild he generated by a naive and 1 1 t i t oilstritilid generation of all tlie clioicw tqreseiited by the choice point) are relevant to tlii\. IIonc1 . the u\e of a utility fiiiiction guided hy task specifiers given to the controller will I ) ( > ( ~~p lo i~c t l later for 0-Plan2 when it is ri\ed i n continuous command and coiit rol applical ion\.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (32)

7 0-Plan2 Architecture

This section describes the 0-Plan2 architecture in detail and describes the iiia jor Iiiodules which make up the system. An agenda based architecture forms the central feature of the system and the design approach. Within this framework, however, the emphasis on and development of search strategies has been concentrated into crisper notions of choice enumeration, choice ordering, choice making and choice processing. This is important as it allow:, us to begin to justifiably isolate functionality which can be described in terms of

0 triggering inechanisms - i . e . what causes the iiiechanisiii to be activated,

0 decision making roles - precisely what type of decision can be made

0 implications for search - has tlie search space been pruned, restrirtctl or further con- strained as far as possible.

0 decision ordering - in what order should we choose be twen tlie altcrnat ive decisions possible.

0 choice ordering - for a decjsiou to be made. nhich of the open choice5 shoiild we adopt.

The main components are:

1. Doinaiii Inforination - tlie inforination which describes an applicatioti tlolililin i i l l d tasks in that domain to the planner.

2. Plan State - the emerging plan t o carr! out identified tasks.

3 . Knowledge Sources - the proct5:,iitg capabilities of the planner (p1ou / ~ ~ o d i , f i u / t i o / ~ o p ~ m - tors).

4. Support Modules - fuiictioiis and comti~aint niaiiagers which support thc procebsiiig ca- pabilities of the planner and it 5 components.

.5. Controller - the decision nialier on the order. in n-hich processing is doiio.

A geiieralised picture of the archittcture illustrated with the component 5 to spccialise the architecture to be a planning agent i \ :,lion-n in Figure 2. More detail of each coinponent follows in subsequent sections. Illustr~ition:, of the contents of the niain coiiipoiietit5 itre drawn by referring to a planning agent.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (33)

P L A N STATE

PLAN NETWORK

T O M E

G O S T

RES 0 U RC E a U S A G E

T I M E \VINDOWS

I

INPUT EVENTS

c

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (34)

7.1 Domain Iiiformatioii

Domain descriptions are supplied to 0-Plan2 in a structured language, which is compiled into the internal data structures t o be used during planning. Tlie description includes details of

1. activities which ca,n be performed in the domain.

2.‘ information about the environment and the objects in it.

3. task descriptions t o describe the planning requirements.

The structured language (we call it Task Formalism or TF) is tlie means throngh which a doinain writer or domain expert can supply the doiliain specific iiiforination to the O-Plan2 system, which itself is a domain indepc n d c u t planner. 0-Plan2 eiiibodiei many sei~~.tlt %pace pruning mechanisms using this domain infoi~iriatioii (strong methods) and will fall I)acli oti other weak (search) methods, if these fail. The Ib\li Foriiialimi ih tlie iiiechanisiii tliat eiiablcs tlie user of tlie system to supply domain depeiitlcnt l<iion.leclge to ahhist the systeiii in i t> w a r ( Ii.

7.2 Plan State

In contrast to the infrequently changing cloniain information outlined almvc, t lie plan state (on the left of Figure 2) is the dyiialiiic da ta btructure used during plaiiiiiitg a n d hoiisei the emerging plan. There are many component^ to this stiucture. the prilicipill one> being:

0 the plan network itself. O-l’Iaii2 ha> ietaiiied a partially ordered net\r,orli of activities as tlie basis of its plan repicsentation. originally suggested i n the N O A H plaiiner. In 0-Plan2 tlie plan information i5 concentrated in the “Associated Data Striictiire” ( A D S ) .

The A D S is a list of node ant1 l i t i k 5tructuret noting temporal and rehoiirce information, plan information and a plan hi5tor).

0 the plan causal structure (soinctimes called tlie tFlcdogy) of tlie plan. Rorroning from Noiiliii and 0 -P lan l , tlie tetli lieells explicit information to bbesplain“ \<.liy t l i ~ plan is built the way it is. This rationale i5 called the Goal Structure (G0s.r) a n d . aloiig with the Table of Multiple Effects (TON r ) . provide5 efficient support to the condition achieveiiient support module (Question .\n\n c w i or Q % ) used in 0-Plan2 ( r.f. ( ‘Ilaptliil ti‘s Modal Truth Criteria [SI).

0 tlie agenda list(s). 0-Plan2 s ta r t> nitli a coiiiplete plan. but one wliicli is h*lian-ect”. hence preventing tlie plan from bc3iiig ( i ~ p a b l ~ of execution. Tlie iiatiirc of t lie flatcs present will he varied, from actions n hich n i c nt a higher level than tliat n h i c h the agent can operate, t o notes of linkages i i w ” < ~ r ~ . in tlie plan to resolve conflict . *.I.‘lil\~>’‘ may also represent potentially beneficial. l)nt a5 )et nnproceswd. inforiiiatioir . ‘ l ’ l i ~ agcnda lists are the repository for this i i i f o t ination 11 hich iiiii5t be pi0 (4 iii order t o attain an esecutable plan. The original 0-Plan I iisecl 3 agenda lists. In O-Plan‘L. effort has been made t o improve tlie structiiio of agciitla inforuiation a n d t h c ~ tr igg~iing nicclianisms. Only one iiiaiii ageiida is I<cpt i t i i ) I d i r \ tat(’ although i ~ l t e r i i a t i ~ . ~ platr \ t i11(’ \ \till require a separate agenda as in 0-PI~i t1 I .

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (35)

w

The plan state is a self-contained snal~shot of the state of the planiiiiig system a t a particular point in time in the plan generation piocesi. It contaiils all the state of the system hence the generation process can be suspendcd ant1 this single structure rolled back at a later point in time to allow resumption of the search'.

7.3 Knowledge Sources

These are the processing units associated with the processing of the Aaws contained in the plan and they embody the planning knowlcclge of tlie system. There are as many knowledge sources (KSS) as there are flaw types, includiiig the interface to the user wishing to exert an iiifluence on the plan generation process. The I<SS draw on information from tlie static data (e.g. the use of an action scheina for piirpohe5 of expaii~ion) to proces a single flaw. and in turn they can add structure to any part of the plad \late ( ( . g . adding htructure to the plaii. inserting new effects or further populating the agriicla(.,) u i t h flaws). \

7.4 Support Modules

In order to efficiently support the iiiai i i planning functionality and provide constraint manage- ment in 0-Plan2 there are a niiiiil~ci of \upport modules separated out from the core of the planner. These modules have carefull! designed functional interfaces in order that we caii both build the planner in a piecewise fa51iioii. aiid in particular that me can experinient with and easily integrate new i~iipleiiieiitatioii\ of the modiiles. The modularity is posihle only through the experience gained in earlier plaiiriing projec:\ where support function requirements were carefully separated out froin the general prolilein \ol\riiig and decisioii making demands of the

Support modules are intended to pro\ itle efficient support to a higher level where decisions are taken. They should not take any dcc ition theinsel~.es. The!- are intended to provide complete iiiforinatioii about the constraints the! are managing or t o respond to question., being asked of thein to the decision iiiakiiig level itxelf.

The support inodules include the f o l l o ~ ~ jng:

0 Time Point Network ( T P N ) lIi\11:ig<\1' t o iiiaiiage iiietric and relative time constraints in a plan.

0 Question-Answering ( Q A ) . Xliiir t o ( 'hapman's hIodal Truth Criteria [ 8 ] , this is the process a t tlie lieart of O-Plan2's condit ion aclijc~.emeiit procedure. It answers the basic question of whether a proposition is trric or not a t a particular point in the plan. The answer it returns may be ( i ) a categorical * * j e i * * . ( i i ) a categorical bb~io". or ( i i i ) a "maybe", in which case Q 2 will s u p p l ~ . an ~111(~7iaii1 >ci ( ~ t r u c t u r e d as a tree) of 5trategies which a l i ~ i ~ ~ l ~ d g e hoiirce can c l i o o ~ ~ f r o i n 111 order to enhiire tlie truth of thc piopo\ition.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (36)

e

e

7.5

TOME and GOST Managemrnt ( T ( ; M ) to iiiaiiage the causal structiiix) (conditions and effects which satisfy them) i n a plaii.

Plan State Variables Manager to iiiaiiagc partially bound objects in t h e plan.

Resource Utilisation Maiiagrnicnt to nionitor and manage tlie uhe of rehoiirces i n a plan.

Instrumentation and Diagnostics routines. 0-Plan2 has a set of routines which a l l o ~ the developer to set and alter levels of diagnostic reporting within the system. These can range from full trace information to fatal errors only. The instruinentation routines allow performance characteristics t b be gatliered while the system is running. Information such as how often a routine is accessed, time taken to process a n agenda entry, etc, can be gathered.

Controller

Holding the loosely coupled 0-Plan2 fr;\uleworli together is the C'ontrollcr acting on the ageti- das. Items on the agendas (the f l a w ) n i l 1 haire a contest depeiident priority ivhicli the controller can re-compute, and which allows for t lie opportunism required to drive plan generation. The agenda mechanism and manager have heen 4iiiplified from the 0-Plan I work i n that two of the three agendas have been collapsed into a single structure. Entries on this single structure employ a triggering niechanisin for activating the 1tno.ivleclge sources via the u s e of plan state or other data. Triggering on specific occur1 ('11ccs. 5uch as tlic binding of a variahlc. the satisfaction of a condition, the occurrence of an estcrnal e\.ent, etc.. allow an eficiency to be Iiuilt into 0- Plan2 that was missing in 0-Plan1 . I\ hich u s e d a priority hcheme whereby agenda entries were prioritised at time of entry. This enhancctl scheme does have an impact on the est ra coiiiplesity of knowledge source required, forcing nile> to be s ~ t regarding the writing of kno~vlcclge sources. In return however, this has given 11s lilio\vledgc~ soiirces with niucli greater capability than pre- viously achieved. For example a knoi~ Ic~lge 5ourcc ma!. be able to dynamically create a trigger for the continuation of another agenda entrj. oil su5pensioii of tlie current ent

An agenda of alternative plan state5 i5 alho held by the C'ontroller for search purpohes as was the case in 0-PlaiiI.

7.6 Discussion

Having reviewed the main compoucnt~ i n tlrc 0-Plan2 architecture, we wihli to m a k e some observations on a number of issueh.

7.6.1 Knowledge Sources

The 0-Plan1 planning prototype allovxd Ii~iou ledge source> to perform opcratioiih at a rcdatively low level. This proved unsuitahlc for b o i i i ( \ planning aclivitieh. such as that of cspancling a n action, or satisfying a condition. I V ~ P I Y ~ thcrc is gcirrrall~~ a large amount of \vork involved. This includes the iiitroduction of strncturo to i 11(. pldtl < 1 1 1 d t l i c Imling of elFcct5 aiitl coiitlitions. -411

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (37)

these entities are related - so 0-Plan I had difficulty treating these sub-operationh as separate schedulable agenda entries with suitable priorities. In the later stages of that research we introduced the notion of a sequence to re-establish the relationships between the various entries, with partial success. A cleaner mechanism, which we refer t o as compouiid agenda entries, is being explored for 0-Plan2 to allow for linomledge of complex sequencing of planning decisions t o be provided to the planner by the knowledge source writer 1421.

In addition, 0-Plan2 employs a knowledge source staging scheme where tlie knowledge sources allow for work to be deliberately stngfd [42]. At each stage the information within the agenda entry is progressively built up for u s e in later stages. Oiily the later stages are allowed per- mission t o alter the filial destination of thi? iiiforiiiatioii - the plan state. At the end of each stage the knowledge source needs to satisfy stcigiiig coiiditioiis in order to continue processing to siihsequent stages, thus the cont lollrr. ha5 tlie ability to halt processing and siispeiid the knowledge source. The agenda I(’( 01 (I i t >elf ( ies all the ”state” of the procehsing, so caii safely be retnrned to the agendns foi Idtor rc~itinptio~i: the knowledge soiitcw tlicwiselves are st at cl ess.

The advantages of this scheme ai.(’ iiidiiy: fir\tl> there is no longer the yes/no situation of whether an agenda entry caii be pro( oswrl a5 the iiiforiiiatioii can he built up in stageh. This in tnrn offers the controller greater fl(~xihi1ity i n i t s job of dynamically computing priorities for agentla records awaiting processing. l‘liis i i i u c l i eiihaiices the ability to exploit parallelism and opportunism in tlie system.

Iinowledge sources run on Iiiiowlctlgc~ Soul ( e Plat foriiih. wliicli are basically procehhiiig engines for the knowledge source code. The eventual 0-Plan2 will exploit multiproceshor architectures, where possible, so tlie current systeni l i a s a clean separation of its knowletlge hource platforms from the other system modules. and loc.liiiig iiiecliaiiisiii~ will be put in place to ensure that data in the system is up to date and conistCiit. Only the final stages of a l<nowledge source can change any of the plan state: car1ic.r \tag(>\ iiicwly h i l d up information locally. \T’e intend to investigate a language for de\cril,iirg Iiiion-ledge Sourceh (Iinowledge Source Framework). Amongst other tliiiigs this mill allon foi iiifoi mat ion concerning the selective lochiiig of parts of the database to be gathered.

7.6.2 Controller Strategies

The Controller plays a major rolr i i i t ) IC> 0 l ) C i ’ i i t ioii of‘ the planner, aiid is largely rehponsihle for acliieving the degree of opportiiuihui wii1:Iit i i i O-Plaii2. Its 11iaiii role is to clioohe a candidate from amongst the set of currently I i.iggc.iwl <lg(’li(lil cii t ric\ to lie loaded onto an appropriate and available kiiowlcdge source platforiii. I ‘ O I t h i \ I’(’rl\Oll t hc ( ‘ontroller is also 1;iion.n ah the Agenda IIaiiager. In order to do this woi.li (~ f Ioc t i \ el! r~ i i ( I flexil)lj. the controller i i i i i y t coii\ider priorities attached to or computed for each ol‘ t l i v t riggcied dgeiitla entries. Prioritiey can be relatively coniplex aiicl based around tlic f yp r of t IIV agc>iitla mt and its iiieasiire of cleterniiiiism. 0- Plaiil used lieuribtic measures tlctiiiliiig l l i c aiitount of choice contained i n an eiit the “top” (i.e. the measure of choice WCII iiitiiietliat~ly ) and at the “1)ottoiii“ (i.e. a measure, or wtiiiiate, of the eventual clioicc. C I I ( or i i i t t . r td i f t l ic entry is chosen). In 0-Plan1 these were rcfciwtl to as the Biwnch-2 estiliiiltol ( t )I(’ i i i i i i i o ( l i r i t C I)rancliiiig ratio for tlir choice point) and

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (38)

the Branch-n estimator (a heuristic chtiiiiate of the number of distinct alternatives that could be generated by a naive and unconstrained generation of all the choices represented by the choice point). These measures have proved uhefiil in distinguishing betweni choice items aiid they ensure that opportunism is exploited where po55ible.

The controller is designed in such a xvay that it can operate with different pre-loaded strategies and utility functions. At present tlie hystem operates with a simple default strategy (knowledge sources priorities fixed by the user) but as the representational range of the Task Formalism increases it can facilitate the loading of domain specific and specialised strategies and utility functions. The controller will be the subject of further research as we wish to tlevclop more powerful strategies, including:

0 Qualitative Modelling. As 0-Plari2 del ~ l o p 5 for iise in continuous coiii niand aiid control applications the need to predict ant1 reroi.ci from 4tuations become5 111ur11 inorc demand- ing. An important role for the (.ontroller then i \ to behave in a uiiich morc pro-active manner, exploiting as much kno\vledge of the eiii as possible. Thc oar1it.r work of Drabble [13] provides a good starting position for how this \vi11 be acliio\-rtl.

0 Ordering Mechanisms. Temporal (’olierence (T( ) [1.51 showed that algorithins must be developed to address the many variants of ordering problenis ( TC addreswd tlie problem of “condition” pre-ordering). Effcrti Ire coiitrollcr operation requires recognition of triggering mechanisms for appropriate 01 tlcri 11:: r t la tcd algoritlrms.

7.7 Process Structure of the 0-Plan2 Impleiiientation

The current architecture is able to siipport both a planner and a simple eseciitioii agmit. The job assignment function is providcd by a scyarate procesh which has a sjniplr I I I ( ~ J ~ I I j n t c~face.

The abstract architecture described iri Figure 2 can be mapped to the 5ybtc.ni and process architecture detailed in Figure 3 . (.‘oiiiitiunicatioii hetn-eeii the various procesw\ a n d managers in the system is shown. Each entry Ivit l i i i i the Figure i \ explained later in this w-tioii.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (39)

L E F T I N - 4-

L E F T O U T

T R IGG ERYIN

A G E N l3.i I N

I 1 Ij I N

Diary

Data base Man ager

I Trigger Detector 1

Plan State

Agenda Table

R I G H T O U T .L

- RIGHTIN

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (40)

7.8

The hasic processing cycle of the Ijl i i i i iIt’ i. is it\ follon~s:

Processing Cycle in the Current 0-Plan2 Systeiii

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (41)

with a given time. For esaitrplc. w i i 0 a( t i o i i 3.2 for execution a t 1:02 01‘ trigger a visit bank activity at 930.

Eventually tlie agenda entry i s selected for processing by tlie C‘ontroller/Agcntla LIaiiager.

-2. The Controller/Ageiida Manager assigns an available Iinowledge Source Plat forin ( K P ) which can run the pre-nomiiiaictl Iinoivledge Source on tlie triggered agenda eiit ry.

5. When a Knowledge Source Plat forin h a \ heen allocated. if it does not already contain the nominated Knowledge Source. the Platform may request the hody of the Iiiiawledge Source from the Database hhnager . in order to process the agenda entry. Iinowledge Sources may be stored with tlic Platforin SO this request is not necessary in a11 cases. Some platforms may be best siiitcd to run particular knowledge sources, hence the system will not store all knowledge \OIII‘CC\ at all platforms. The knon-ledge source I)liitfor11is will eventually have their own local Iil)rar of knon.leclge sources. Loclii ng tlo~vir of a specific real time knowledge source to a t l d ( < i t ( ~ I platforin i h dloived for in tlic clcsigii.

6. A protocol (called the Iinon-lcdgc S o i m c 1’i.otocol) for communicatioit h t \\“W the con- troller/agenda manager and i~ I ~ ~ I O N ledge. ~ o ~ i r c e rnnning on a plat forin cont r o l s the pro- cessing which the knowledge W I I I Y ~ ( ~ (’<in do ant1 tlir accehs it has to thc ( n r r c n t plan state via the Database Manager (1111 ). -1 linon Icxlgc3 boiirce can ternliililt e \vi t 11 none. one or multiple alternative results tlrroiigli iiitt.riiction n-ith the Controller via t h i b protocol. The Con troller uses an A1 t e r m t i ves 1 I ii i i age. r S 11 pport Alodule to act 11 a 11). I I I a i I ii gc’ a 11 y alter- natives it is provided with illitl to \ c ~ ~ ~ l i altcwiatires n-lien no result\ arc’ wiiirired I)!* a knowledge source. A knowletlg(~ w u i w ( <iii i l l 5 0 I I P dbl<etl to tcriiliiiatc a t i t h itcst “htage” boundary b y tlie controller.

The internal details of the Database l I < i iiager ( r j u ) will depend npoii the particular rcpresenta- tioii chosen for the Plan State. In I~’ig~itx~ :1 t l i r iirtciriial details of the I la tahaw iriaiiager relate t o the 0-Plan2 planner. Here there i5 ii wprdtioii of t h c .l>soriatetl Data Striictit r c ( . ID\) level which describes the plan networli. i I i ( L ~ l ’ i l l ) l ( ~ of 1111It ipl t~ Effects ( T O \ I E ) antl t l i c . Goal Structure Table (GOST) from the lower l e ~ e l ’I i l l ) ( \ Point S(\I\\o~li (TPY) antl i t 5 ii\\ociiit(>(l iiiotric time point list called tlie Landmarli Lino ( I I ) .

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (42)

8 0-Plan2 Planner

8.1 Plan State

Tlie planning agent plan state holds illformation about decisions taken during planning a,nd information about decisions which are still t o he nia,de (in tlie form of a.n agenda).

8.1.1 Plan Network - ADS and T P x

The Associated Data Structure ADS provides the cotitea'tuctliiiforinatioii used to attach meaning to the contents of the Time Point Network T P N . and the data defining the emerging plan. The main elements of the plan are activity. dummy and event nodes with ordering information in the form of links as necessary to definc tlie partial order relationships bttn-eeii thwe elements. Tlie separation of the A D S level froill the time points associated with the plan entities is a design feature of 0-Plan2 and cliffus froni our previous approach in Noulin and 0 - P l a n t . It is motivated by our approach to time point con5traint inanageinent [Ci] ~vhich reasoiih about both ends of plan entities (such as nodes aiitl link\) nnd n hicli can lie more efficiently iinpleinented where there is uniformity of representation.

Time windows play an important part in 0-Plan2 in two mays: firstly as a lllt'illls of recording time limits on the start and finish of ail action and on ith duration and delays Iwtweeii actions, and secondly during the planning pliahc itself as a ineaiis of pruning the potential search space if temporal validity is threatened. Timc jvindows in 0-Plan2 are maintained as min/iiiax pairs, specifying the upper and lower l~oiiiid\ linon~i a t the time. Such bounds niay be ~ynilmlically defined, but 0-Plan2 maintains a iiuincrical pair of bounds for all such nuiiicrical values. In fact, a third entry is associated with such iiiinierical bounds. This third entry is a projected value (which could be a simple number or a more complex function, data structui,e, etc.) used by the planner for heuristic estimation. h c ~ ~ l i control and other purposes. ' Higher level support modules (such a5 QX. t h e T O M E and G O ~ T Manager. ctc.) rely on the detail held in the A D S and on the fiinctionalitj. provided by tlie T P N . The ID< is niaintaiiied by a set of routines which we refer to the Setnorli AIanager.

8.1.2 TOME and GOST

The Table of Multiple Effects ( T O Y I F ) Itolds htatetiients of form:

( fn a r g l arg2 . . .> = value at time-point

The Goal Striictiire Table (GOC;T) l i o l t l 5 \ t , i tc i i rc i i r t 5 of r'orrii:

condition-type ( f n a rg l arg2 . . . > = value at time-point from contributor-list

'AI1 numerical values in O-PIaii,i ale LicIci < I - -LI( I I 1 i i p I c -

. j o

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (43)

where contributor-list is a set of pairs of format: (time-point . method-of-satisfaction-of-condition)

In the current implementation. effects and conditions are kept in a simple pattern directed lookup table as in Nonlin [39]. The 0-Plan1 C'/oud.\ mechanism [A13 for efficiently inaiiipiilat8iiig large numbers of effects and their relat iondiip to supporting conditions will he uhed in 0-Plan2 in the course.

8.1.3 P l a n State Variables

0-Plan2 can keep restrictions on p lan s t ate \ aiicihlc\ \\ itlioiit necesharily iii\i\ting that n definite binding is chosen as soon as the vi:ri;il)lcl is i i i t iocluced to the Plan State.

8.1.4 R e s o u r c e Util isatioii T a b l e

The Resource Utilisation Table hold> S I ; i ~ c i i i r n t b of foriii:

set/+/- (resource (resource-name> <qualifier> . . . I = <value> at <t ime-point >

The statement declares that the pai~ticiilai~ ~ n o i i r c e is set to a specific valne or changed by liciiig increinented or decremented h y 1 l i e giveii aluc at the indicated time point. There can lie uncertainty i n one or both of the \ .iiItit. atlid the tinic point which are he ld ab ~iiiii/nlax pairs. 3

Task Forinalisin reyource usage \pecific rltioiis oil r l t tion\ are used to ensure that re\ource usage iii a plan stays within the boui id~ i n t l i c - c i t c d . 'I'lic~e are t n o types of resource usage statements i n TF . One gives a spwiJSccttioti of thc overall liiiiitation oil rebource umge for ail activity (over the total time that the activity and r>i i> (' on of it can span) . The other type deicribes actual resource utiliscrtion at poilit\ i l l t l i e i i o n of a action. It innst 1~ pos4ble (xvitliin the min/max flexibility in the actual ~ P S O I I I Y e ii~agc' statements) for a poiiit i i i t h e ia ige of the siiiii of the resource usage statement\ I O 1)c \ \ i t h i l i the overall hpccificatioii given. 'I'hc Resource Utilisation Table inanages the actliiil r(w)itl~ ( P i i t i l iscit ioii at points in the 1)Iiiii.

8.1.5 A g e n d a

.I 1

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (44)

A n alternatives agenda of plan stat(>% other tliaii t lie current one, which call he considered if this plan state is unsuitable to achieve the tij:,k i:, kcpt by the Controller via the Alternatives Manager Support Module. Formally, all possible Plan States known to the alternatives manager, including the current plan state should be coii:,idered as the "state" of the agent.

8.2 Planning Knowledge Sources

The 0-Plan2 architecture is speciali:,etl into a planning agent by including a numher of knowl- edge sources which can alter the Plan State i n various ways. The planning knowledge sources provide a collection of plan moclijurfion oylatorx which define the functionality of the planning ggent beyond its default 0-Plan2 arclrit cctu ro properties ( essentially limited to coimiu nication capabilities by default ).

The planning knowledge source:, i n t l i v (iirrciit 1 cr4oii of the 0-Plan2 plan tior arc:

KSSET-TASK a knowletlgc~ 5011r( (1 to w t irp a n initial plan stale (or~t~c~s~~otitling to the task request from the job assign t r i ~ t i t aget i t .

IiSXXPAND a knowledge soiii'c(> to osp;ititl a high level activity to l o n ~ r lcvcls of detail.

IiS-CONDITION a knowledge soiit(c t o c'nsitic thnt certain type5 of contlition (only uiisu- pervised ciirreiitly) are satisfictl. l'lri:, i \ i i o i i tiall~. po5ted by a higher 1ci.d IiS-E:SPA4ND.

TiSACXIEVE a knowledge so111 (Y i i i i t i a t c d 1)y IiS-I:SP.\ND fot aclii(>r(i conclit ions.

KS-OR a knowledge source to \('I(>( t oiip of \et of possible altcrnativr liiiliiiig:, and plan state variable hindings. The \et of'nltcl.nilti\'c liriliings and lintlings Ivill Iiavc h e i i created by other knowledge source:, (suclr a \ KS-( 'OX l ~ ~ r I ' l O Y ) earlier.

KSBIND a knowledge so~irce i i s t ~ l to \elect a binding for a plan Stale v a r i a h l ~ i n circum- st aiices where alteriia t ive po<si I)lc I)i 11 diu g s iwiia in possible.

KSl'OISON-STATE a knonlctlgc~ ~ O I I I Y ( ~ I I W I t o tltlill with a statenreti1 l)j, aiiother knowl- edge soiirce that the plan statc i5 iiicoii\i\tctit i n \Oll ie nay or cannot Ic<~ltl to a valid plan ( a s far as that linowledge soiir(c i \ ~ ~ u m ~ ) .

IiS-IJSER a knowledge s o i i r c ~ [it t i \ x t c ( I i I t t he request of the iiser act itig in the role of supporting the planning prow\\ ( I'laiiiicr I - \c r Rolc). This i:, used a t prcscnt to provide a menu to browse 011 the l l l i l l l \ t i l l ( \ ntid l ) o t ( ~ l i t i i \ l l j ~ I O d t c r the priorit!. of hotlie choices.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (45)

0 KSDOMAIN Call the Domaiii Inforiiiation (iioriiially T F ) compiler to alter the Domain Iiiformation available to the agent.

0 I<S_EXTRhCT_RIC,HT Extract a plan patch for passing to the subordinate agent to the ‘right’ of this agent - i.e the execution agent. In fact, in tlie curreut impleiiientatioii, a knowledge source with name T<S_ES€?(‘ITTE packages a plan for execution and then passes this to I<S_EXTRACT-RTC~IIT for comniuiiication to tlie execution agent.

0 I<S_EXTRACTJ,EFT Extract a plan patch for passing to the superior agent to the ‘left’ of this agent - i.e the job assignment agent. In fact this communication between the planner and the job assigner in the current impleiiientatioii is performed by two knowledge sources. I<S-EXTRACT is used to pas3 reqiieited information (such as when iiiforinatioii about a plan state is requested by the ii\cr) back to the job a5sigiimeiit agent (or to a plan or world viewer process as appropriate). I<S-l’T,.\?;?;€~R_FINISHED is used to inform the job assigninelit process that the p l a i i n c ~ ~ ~ l ia i coinplcted it, task.

0 I<S-PATCH Merges a plan patch oii an inpiit event channel into the current plan state. In fact, in the current implenie~tt;ii ioii. t h c w i \ 110 u\e made of I<S-I‘-IT(’H directly.

It is intended that coriiiiiniiication bet~reeii the three agents in the O-Plan2 system (job as- signer, planner aiid executioii systeiii ) \vi11 1 cs11e( t the pliilohophy of coiiiiiiuiiicatioii via plan pat ches and t 11 a t the I< S -EXTRA C’T-1, 1- 17‘1 . I< S -E:S TK -4 C‘T-RIGHT an d I\: S -PAT C H knowl- edge sources will lie the oiJy oiics n-hi( 11 \vi11 i i i a k ~ u\e of the event cliaiincls directly.

8.3 Use of Coiistraiiit Managers to Maintain Plan Iiiforiiiation

O-Plan:! uses a number of con.strnt,,t t i ~ u ~ ~ o y r s to inaiiitain information about a plan while it is being generated. The inforiiiatioit can tlieii be iitili\ed to prune search (where plans are found to he invalid as a result of pi~opagatiiig t h ~ l coii\traints managed by t h e managers) or to order search alternatives according t 0 \om(’ he~ir ic priority. These nianagerb are provided as a collection of support nioduIts nliicli c ai1 I) ( \ callccl 1)). lxioivledge sources to maintain plan information.

8.3.1 Time Point Network Maiiager ( T P X \ I )

0-€’laii2 uses a point based temporal rel)t went <\tion v i t li range constraints lietween tiiiie points and x+itli the possibility of specifying raiigc (oii\iraiiit\ ielative to a fixed tinic point (tiiiie zero). This provides the capability of specif! iiig iehtive aiid metric tiiiie constraints 011 time points. The fuiictioiial interface to the l’ i i i i r l’oiiit Sc tuo rk ( I PS). as 5ec.n liy tlir .\\\ociated Data Structure ( . ~ D s ) has no depeiideiicc oti < I ])ai t i c nlar icpir\eiitatioii of the phii . ror example, rather than the siiiiplc ‘before‘ relation\liip i i w d in tlic O-Plan2 planner’s plan \ t a t c representa- tion. a parallcl project exploring t c ~ t i r l ) o t <11 logic \. it>a\oiiing mecliaiii~iii~ atid i.t.l””~“i’tatioiis for planniiig is investigating alteriiati\c Iiiglioi I ( > \ ( ’ 1 .\\wci<itwl Tlaln Siriictiirc~ i i i i i r rclationdiips.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (46)

The Time Point Network is the lowcst level of temporal data structure and consists of a set of points (and associated time ro1Istriiiltts) each of \r.hicli has an upper and lower hound on its temporal distance from:

1. other points in tlie network

2. a (user defined absolute) start time referonce point

The points held in the TPN may be intlirectl;\7 associated with actions, links and events, with the association being made at the Associated Data Structure level. The points are numbered to give an index with a constant retrieval time for any number of points. This structure allosvs points t o be retrieved and compared through a suitable iiiodule interface and with a minimum of overhead. The interface is iniportaii~ a n d reflecth tlie fzcizcfio72crlity required of the T P N , and hides the detail. This ensures that we have no aho lu te reliance on points as a necessary underlying represen tation. Time point\ ~vliose upper and lower values has coa\-erged t o a single value are inserted into a time ordered I,andn~arli Line ( L L ) . This allows the planner to quickly check the order of certain points n-itliiii the plan. The T P U and L L are maintained by the Time Point Network Manager ( T P N M ) . '1s ~ c l l a \ i t \ use in the 0 -P lan2 activity orientated planner, the ciirrent T P N M h a also been applied to large wsource allocation scheduling problems in the TOsC.4 scheduler [7] where the iniinl~t~r of tinic point5 Ivah in excess of 5000 and the number of temporal constraints exceeded 3000.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (47)

Figure 4 and Figure 5 show the use of the T P Y for applications involving task planning and resource a.llocation.

to (0s))

t l t L t 3 t l

Resource I

m s I I I I I

'r 1' N 1 I I I I I TI .. T1 I I -1 2 .. 1-2 I I 0 bo *. -a -a

I I 1 3 t t-l

,

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (48)

8.3.2 TOME/GOST Manager (T(:IrI)

The conflict free addition of effects and conditions into tlie plan is achieved through the TGM,

wliicli relies in turn on support frotii the QA support module which suggests resolutions for potential conflicts.

8.3.3 Resource Utilisation Management ( R U A I )

0-Plan2 uses a Resource IJtilisation LIanager to illonitor resource levels and iitilisation. Re- sources are divided into different types s i i c h a5:

2. Re-usable: these are resources I\ h i ( 11 < I I P i i s c d and then retiirncd t o a cot1itiioii ”pool”. For example, robots. ~vorli1ntw. lor1 ies. et ( .

Consiiinable resources can be suhcatc~gorisctl r i s s l r i c t l y cou.<ur/ird or may he p/odtic(rbIc i n some may. Substitutability of resource:, one for the other is also possible. Some may have a single way mapping such as money for petrol and sotiip can Be two way iiiappings siicli as iiioiiey for travellers’ cheques. Producable ant1 h i i l ) s t i t ii t ~iI ) lc resources arc difficult to t l c ~ i l lvith because they increase the artioiiiit of choice a v ~ i i l a l ) ~ ( ~ \\ i i h i i i a plan and thus op(ii u p t hc :,earth space.

The current 0-Plan2 Resource TTtilisc\ticin hl:\lii\g(’r nse5 the same hclieiiie for strictly coiisiiin- able resources as in the original 0-l’lati 1 . I l o n c ~ i e r . a new scheine based 011 the niaintenance of optimistic and pessimistic resour( P pi ofilcs \i i t 1 1 resource iihage events a n t 1 a ( tivitie:, tied to changes in the profiles is now i i n t l r r s i i t ( 1 ~ .

8.3.4 Plan State Variables Manager ( P \ I , \ I )

The Plan State Variable Manager is t~csponsil~lc f o r niaintaining the coli~istcwcy of on plan objects during plan generation. 0-]’la112 adopt5 a bah t coiutni t uient approach to object handling in that variahles are only ljoittitl a s citi(l if Iicn uecehsary. For esatriple. i n a block qtacltiug problem, moving block A to Ijloc li H i i i m i i h that it is necessary to cotisider the object which A was previously on top of a n d ftotti \I I i i c l i i t \vas t n o \ ~ ~ l . ’I‘liis otiject is ititroduced as a plan state variable whose value ni l1 I)c I ) o i t i i t l <is <i i i t l \I lieti iieces5at.y. O-l’lan 1 itsetl a separate agenda to hold variable binding agclitlii oiit r i os . l.liis s( 11ciiic provctl to be difficult to uhe due to tlie miiiiber of constraints n l i i ( l i \ \ ( ~ I ( ~ 1 ) t r i l r III) l)o1uwii r ig~~ i ida ciitrieh a t i t l ivii l i i i i agenda entries. The constraints were hpe( i f i t d <is:

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (49)

a Constraint-list: This specific.\ a list of attributeh which the value to which the plan state variable is bound must lia\re. For esaiiiple, it must be green, hairy and over Sft tall.

To overcome these problems a separate Plan S i ate Variables Manager within the Database Manager ( D M ) has been implemented which niaintains a n explicit “model” of the current set of plan state variables (PSV).

When a PSV is created by the planner the Plan State Variables Manager creates a plan state variable name PSVN, plan state variable body P ~ I ’ B and a range list from which a value must be found. For example, the variable could be tlie colour of a spacecraft’s camera filter which could be taken from tlie range (red green blue yellow opaque). A plan state variable must have an enumerable type and thus cannot be. for example, a real number. The PSVB holds the not-sames and constraint-lists antl i? pointed to by one of inore P S V N ~ . This allows easier npdating as new constraints are added and P \ \ R’FI are macle the same. Two or more PS\‘B’S

can be collapsed into a single PSVB if a11 of the constraints are compatible. i.e. tlie not-sames and constraints-list. A PS\” pointing to a collapsed P<\.B is then redirected to point at the remaining PSVB. This scheme i h a lot iiiore flexible than the previous “sanics” scheme as it allows triggers t o be placed on the biiicliiif: of P \ I . * ~ (e.g., do not bind iiiitil tlie choice set is less than 3 ) and allows variables which a i c> crcatjng bot t lc t iech~ to be identified and if necessary further restricted or 13ound.

8.4 Support Mechaiiisiiis in 0-Plan2

-1s well as tlie managers referred to a l ) o \ ~ . i t iiatnber of other support routines are available for call by the I\-nowledge Sources of 0- P l a i d flip iiiain such siipport ineclianiinih which have been built iiito the current 0-Plan2 I ’ l a ~ i ~ i c ~ iiic l i i de :

a Question Answerer ( Q A )

The Question-Answering modiil(~ i ? thc c o ~ e of the planner antl mnst IIP Iiotli efficient and able to account for tenipora\l coiist 1 a i i i t5. Q.4 hupports the planner to satisfy and maintain coiiditioiis in the plan i i i (onflict free fa\hion, suggesting rciiicdies where possible for any interaction5 c l c t r c t ~ d . (2 4 a5 inipleniented in 0-Plan2 is an efficient procedural interpretation of Chapinan’? 1Iodal Truth Criteria [8], \vhich was distilled from Q A in Nonlin [39]. Q A pro\’itie~ siipport for the TGM in the system. a i i t l i > supported in turn by ailother low I ~ v e l niotliilc ( i i r lpl i Opcrations ( G O P )

0 Graph Operations Processor ( (40~)

The GOP is a software iinplcmeti t a t iou ol il grnph procebhor. providing efficient answers to ordering related question^ Ivitliiir t I I O tiiiiiii plan (represented by a graph). G O P works within temporally ordered. a? \\ (.I1 p i t t i i t l l ) oi~tlerod. activities i n tlic gi.apIi.

0 Contexts All data within the 0-Plan2 ] ) l r i i i \trite ( (III 1)c “contest layered“ to provitlc hupport for alternatives nianageinent and (oiitcst I ) ~ i ~ c ~ t l I ectsoniiig. A n eFicient. s i ructiire sharing support modr1le provides t l ic i l l ) i l l i > 1 0 (oittc>ut l a ~ . c ~ i an!. da ta strric.1 IIIYI ilccehhor antl

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (50)

e

updator function in Lisp. This i:, particularl> useful for the underlying content addressable database in the system: O-Base.

O-Base This database support module supports storage and retrieval of eiitit4./relatioaship data with value in context. This iiiodel ajloit-:, for retrieval of partially specified itenis in the database.

Jii addition, there are support modules providing support for the TJser Interface, Diagnostics, Instrumentation, etc., and there are others nrhicli still need further development (e.g.. variable transaction management).

8.5 Alternatives Manager

There is an additioiial support module c,ipability iri O-Plan;! which is utiliserl hy the ('ontroller. This provides support for handling ; ~ l f r i ridtil.(l 1)li i i l :,late\ ivithiii a n O-Plan2 agelit.

If any stage of a I ino~ledge source find\ that i t has alternative ways to achieve its task, and it finds that it caiinot represent all those alteriiativeh in boiii~ n a y within a 4iigle plan state. then the controller provides support t o allon t h e altriiiatives that are generated to he iiianaged. This is dolie by the kiiowlcclge source tclliiig the controller ahout all alternative:, bit1 one favonred one and asking for permission to cotitinue to proc this (by the equivalent of a stage checli). This reflects the 0-Plan2 search stratcgj. of locnl t h c n globrrl b f s t . A support routine is provided by the controller t o allow a l i i i~\~letlgc s o i i i ~ ~ writer to inform t h r coiit roller of all alternatives but the selected one.

A knowledge source which cannot acIric\ (1 its tn:,l, or M hich decides that the curreli t plan state is illegal aiid cannot be used to g e ~ i ~ r i i t e a valid plait ma!. terminate ant1 t ~ l I thc coiitroller to poison the plan state. In t h irrcnt \.crsioii of 0-Plan'L, this will i ior~i i~~l ly initiate con- sitlcratioii of alternative plan st hy a dialogue. hetn-een the controller anti I he alternatives manager. A new current plan state ivil l be selected and beconie visible to n e ~ v l ino~ledge soilice activations. Concurrently running li~io\vlt~lge hou i ces v, orliilig 011 the old (poi\oncd ) plan state will be terminated as soon a:, possi1)lc ( a t t110 t icst htage koiinclary) a:, tlirjr. cffort:, [vi11 be 1t';lS t ea.

8.6 Imp 1 e m e 11 tat io 11 as S e p ai- at e P ro cess es

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (51)

can be processed by the Interface Manager ( t h e iiianager in charge of all interaction, diagnostic handling and instrumentation) as they occur. The reaction time performalice of the systeiii is measured by the time taken to post a n agenda entry by the event manager and it being picked up by the agenda manager once triggered. The cycle time performalice of the systeiii is measured by the reaction time plus the tiine to ahsign the agenda entry to a knowledge souxe and have it run to completion.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (52)

9 0-Plan2 Job Assigner

In tlie current iinplementatioii of 0-Plan2, job assignment is a simple process with a menu of options available to the user. C'omiiiiiiiicatioii between the job assignment agent and the plaiiiiiiig agent of 0-Plan2 does not cnrreiitly reflect our intentions of coiiiiiiiinicatioii via plan patches.

The current menu of choices is:

0 Iiiitialise Planiier

0 Input TF (via pop-up inenii of TF files available)

0 Set Task (via pop-up menu of tahlis available in current TF file)

0 View Plaii

0 View W o ~ I d (a.t noiniiia.t,ed node )

0 Replan

0 Execute Plaii

0 Quit

The job assigiiineiit process maintain5 the \et of open coininand clioice~ dcpencliiig on tlie current status of tlie planiiiiig agent (whether i t has been given domain informatioii. set a specific task or is currently planning or has alreadjv generated a complete plan ).

The planner views tlie job assignment procesh as if it mas a full 0-Plan2 agent and takes requirements and Coi1il:ialids in tlie fori11 of events from the job assigner. ' ~ h c ~ planner also pac1;ages its responses to tlie job assigner in the forin of simplified events.

I 0

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (53)

IO 0-Plan2 Execution System

One of the aims of tlie O-PIan2 project i s to investigate tlie issues involved i n linking an intelligent planner with a remote execution agent. In order t o investigate these issues a version of the 0-Plan2 architecture has been configured to act as an execution agent. To configure 0-Plan2 as an execution agent required a new set of knowledge sources to he defined which allow the system to follou~ a plan rather then generate one.

The present 0-Plan2 execution monitor accepts a “plan fragment” from the planner (this is created through the use of a knowledge source IiS-EXECUTE in the planner) together with a set of monitoring instructions specifyjng lion the actions of tlie plan should he monitored. The plan fragment consists of:

1. the plan specified as a partially-order iietnorli of activities

2. tlie TOME, GOST and temporal inforniation built up during plan generation

3. the attacliinent point t o be used by the execution iiioiiitor

The execution monitoring strategie5 1% hich can lie specified are as followh:

1. inonitor all actions and report tlic ~ i i c c e h h o r failure of their execiitioit

2. monitor specified actions for:

( a ) siiccess or failure during execution

( 1 ) ) resource utilisatioii (usage and rcplenishiiient )

( c ) specified start or completioii tiiiie of an action relative to a givtiii wfer(~nce point (external event, time clocli or plan action)

3 . report only when the ~vliolc plaii f‘ragnic111 ha5 completed executioii

The message is received by the loft input giiaid of the execution agent\‘ 1 ~ ~ ~ ~ 1 1 1 hlaiiager and converted t o an agenda entry. LYIi ta the ageiicia entry is processed it caiiw\ t h e knowledge soiircc IiS-13REAII<UP to be run in thc cscciitioii agent. IiS-BREAIiUI’ tali(>\ t It(’ i i i p u i meshage an tl perfor in s t 11 e foll owi ng two s t e p :

1. creates an agenda entry record for each of the actionh in tlie plan. ‘I‘lic trigger for tlie agenda entry will be the time at lvliicli tlie action should be esecut(~t1. l’lir lii~owledge soiirce I<S-T)ISP.ZTC’FI will 11c iisctl to send an action to the right out c t i an i id for cscciitioii.

I 1

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (54)

The Diary Manager is set lip t o initiate triggers a t the appropriate time. \,T’heii triggered, the agenda entry is added to the triggered agenda list to await the availability of a knowledge source platform on which to run. The information derived from the iiionitoring is then assembled into a “return message” for the planner. The message is accepted as an event through the right input guard of the planner and scheduled as an agenda entry by the agenda manager. The knowledge source KS-WORLD in the plaiiiier is used to analyse the message which is in the form of a plan patch. If there was an execution failure then the patch would also contain a flaw i.e. the reason for the failure. For example, a precondition not met, external event to be removed, action which could not be decomposed. etc. ‘. The planner then integrates the plan fragment into the current plan state and adds the flaw to its list of agenda entries.

The work to date on the execution agent within tlie 0-Plan2 architecture is only at a very simple level and has mostly been coiiceriied ivitli cwuring that the comniunicatioii capabilities are present t o address issues of inter-agent plan fragment passing. Further work to cliaracterise the requirements for and capabilitie:, of a reactive execution agent have been undertaken [31] and an associated research project ih now under.ivay to explore how tlie O-Plaii2 architecture can support these requirements.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (55)

11 0-Plan2 User Interface

1 I. 1 Planner User Iiit erface

A I planning systems are now being I I S C ( I i n reali\tic applications by users who need to have a high level of graphical support t o the planning opciations they are being aided with. In the past, our A I planners have provided custom h i l i l t gi.apIiica1 interfaces emhedtled in the specialist programming environments in which t he plaiiii(>r\ l i a v ~ been implemented. It i4 now important t o provide interfaces to AI planners thal are i n o r e easily used and untlerbtood b y a broader range of users. We have characterised the U\PI' iiiterface to 0-Plan2 as being based on two views supported for the user. The firht i \ a P l n ~ , 1 - / ( I / ' tvliich is used for interaction with a user in planning entity terms (such as the I I W of' P rnvc l la r t s , Gant t charts, resource profiles, etc). The second is the World V 7 z ~ w which l ) r ( v i 1 1 5 (Ioitiiiiii oriciit;itetl~vicn. or 5iiiiiilatioii of what could happen or is happening in t e rm5 of' u 01 It1 b t (1.

\ \

I I

\

m * " * . SETUP

atoCtis D M DISPLfi'l DRfUf EDIT I n w w LAYER: S E T r I W S PLOT IJCS: IJTILIT'I

3 D PlSHffiE

SAL'E:

I

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (56)

Computer Aided Design ( C A D ) pacl.;age\ tiviiiliil)lc 0 1 1 a \vide range of microcomputers and engineering workstations are in witlehpwatl i i s c a t i d will probably be lilloivin to potential planning system users already or will be in use soiiienlrerc in tjieir organisations. There coiild be benefits t o providing an interface to an A I p l a i i i i ( ~ r t lit.oiigli widely available CAT) packages so that the time to learn an interface is rediicctl and a raiigt’ of additional facilities can he provided without additional effort by the impleiiientors of A I p l a i i i i t ~ r x .

Some CAD packages provide facilities to enal)lc tailored iiiterfaces to be created to other pack- ages. One such package is AutoC‘.\D [ 11, [%] - though it is by no iiieaiis iiniqiie in providing this facility. AutoCAD provides AutoLTSP, a \.ariaut of the LISP language, in whicli custoiiiised facilities may be provided [5 ] , [37 ] . Tliis is con\.enieiit for work in interfacing to .41 systems as workers in the A I field are familiar with the LISP language. However. the tcchniques employed would apply whatever the custom i s a t i oil I ii I i gii iig(1 \vas.

We have built an interface to the I<d i i i l ) i i i y , l i \ I pl~iiitiiiig 5ystcwis which is l ) a s (~ I 011 AutoC‘AD. A coiiiplete example of tlie use of t li(> i i i t c t (1 lids 1)rc.ii built for a spacv platform building application. 0-Plan2 Task Fornialisiit Iiii\ I ) ( ~ t i \\ rit t v i i to allow the getieixtioii of plans to build various types of space platform Ivith c.oiiiicv.ti\ it! (oiistraiiith oii the 1iiotl111ch and components. A domain contest display facility lia5 IWOII pio\.itleci t Iiiorigh the use of AritoT,ISP. This allows tlie state of the world followiiig tilt. csw t i t i o i i of‘ i i l i ! ’ iict ioii to I N visualihrtl tlirough XiitoC’AD. Means to record and replay visual hit111iI i i t i o i i w i i i o i i ( ( ~ s f o r plan execution ai.(’ pro\,idcd.

X sample screen hiage is incliidc~tl i n 17igui.o ( j . ’ l ’ l i(~i~(~ art’ t I i r t ~ ~ iiiaiti \viiitlo\vs. The planner is running in tlie window to the t o [ ) 1 ( ~ 1 i I i r i i i ( l (o i i ior <it id is hlioivittg i t 5 iiiaiti iiher menu. The planner is being used 011 a \1)ii(’(’ 5 t < i t i t i i i i i \ \ (~ i i i l ) l>~ t k i 5 l i nntl h i i s jiihi 1)oc~ii r ihe t l to get a resulting plan network. In the 1’1u1/ I / ( t i ’ \ul)i)ort ~ ( 1 I)y 0-Plan2, thih has lwen displayed using the Lond Plan menu item i n tlic I ~ I I ~ C . \ i i t o ( ’ . \ l ) \\~ititlo\v along tlic hottoill of the screen. Via interaction with the menu i n t l ic . \ t i t o ( ’.\ 1) iiiilwv. tlie planner has I)cvi i informed that the user is interested in tlie contcst i i i n l ) c i i i i t * 1 i l i i i poiiit i i i the plan - t l i r selected node is highlighted in the main plan dihpla~.. lii t l t ~ 11 odd I ’ i t srippot’ted by 0 - P l a n 2 . the planner has then provided output which ciiii I)c ~isiiiiliwd 1)). a h i i i t able tloiiiaiu spwific interpreter. Tliis is shown in tlie window to the t o ] ) riglit I i i i i id (oii~c’r of‘ the scrceii n1iei.c plait. c4evation a n d perspective images of tlie space s l a t ioii r i r o 5 i i i i r i I t i1 l i~~o\ i>l~ displajwl.

The 0-Plan2 Plan View and Il’orltl I*ieu 5iiI)l)oi t i i i ~ I idii isi i i5 arc tlesigiictl t o re ta in intlepen- deiice of the actual implementations for t l iv \ ivu(’i.5 t l i t ~ i i i w l ~ ~ ~ s . ‘l’liis allowh uritlely available tools like AutoC‘XD to be eniployc~tl \\.lirrc ‘i l) l) iopiiat(>. 1)iit a150 a1101vh text 1). specific viewers to lie interfaced \vi1 Iioiii (.Iiiitigc’ to ( l- l ’ l i i i i2 itself. The bpecific viciverb to be iised for a domain and tlie level of i i i t o i l < i (o ~ I I P \ ( < i i t 5iil)poi.t for O-I’laii2 I I ~ C i h described to 0-Plan2 via the domain Task I~oivii~jli5tii ( I I ) . \ s i i i ~ i l l t i~ i i i i l ) c i~ of r i c w r rhiirnftcri.sfir.s can hc stated. These are supporietl I)! O-l’l,ii i2 <III(I (oiiiiiiiiiii(.iitioiis I;ingii;-igc~ is provided such that plan and world viewer5 caii iiil)iii 1 ( 1 O - I ) I < I I I ~ r i t i i l t r i l \ o o i i t p i i t f i o n i it.

Sophisticated Plan and ‘Il.’orltl 17io\ \ ( i i5 (oit I(1 I)(’ IIWI i i i l ‘ i i t i i i v \$it I i O-l’laii2. \\-e believe that tiine-phased tactical inapping d i s p l i i > \ of t I t ( 1 I \ I)(> iiw(1 i i i i i i i I i t < i r j logi5ti(.s (‘ai1 I)(> iiwd as a R’orld Viewer. \Ye have also coii5i(lvi(s(l i t i l ( 1 i I < i ( ( % \ 1 0 \ . i t i iiiil Iioality o i i \ , i i o i i t i i ( l i i t \\(’ term PlanIVorld-VR.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (57)

11.2 System Developer Interface

When 0-Plan2 is being used by a developcr. i t i> u\nal to have a number of windows active to show the processing going on in the inajor cmiipoiients of the planner. There is a sinal1 window acting as the job assignment agent wi th i t > main 0-Plan2 menu. There are then separate windows for the Interface hlaiiager ( 111 ) - tlirough which the user can coiimiuiiicate with other processes and through which diagnostic airtl iiistruinentation levels can be changed. The Agenda Manager/Controller ( A M ). the Databdbe hfaiiager ( D M ) and the Knowledge Source Platform(s) ( K P ) then have their owii n.intlon h. Further pop-up windows are provided when viewing the plan state graphically or when ge~t ing detail of parts of the plan, etc.

A sample developer screen image is slion-ii i n f igii i~ 7.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (58)

11.3 0-Plan2 User Roles

User interaction with 0-Plan2 cat1 o c ( ~ i t i ~ lor [I \.iiri(.ty of piiiyoses. Varioith I*o/~ . \ of a user interacting with 0-Plan2 are tlefitictl i i t i t l i i t ~ \iii)portod i n tlilfereiit wayh \vi1 h i l i the hysteiii. We consider the identification of t l i c ( l i f l k ~ i ~ ( ~ i i t r o I ( ~ \ to I)e a riscfiil iiitl to gititle flit i i re tiher

i 11 t erface support provi sioii.

11.3.1 Domain Expert Role

i\ single user responsible for clefiiiitig t IIC I )o t i i i t l s oit t Iic appliration area for wliich the system will act. The domain expert user t i i a j ~ t l irct t ly o r i t i ( l i t w t 1 , v hpecify O-PIaii2 'I'ahk Fot.iiialism to define the domain iiiforinatioti \i I i i ( 1 1 t l i t , i ) I r ~ t ~ t ~ o i vi11 IIW.

11.3.2 Doinain Specialist Role

11.3.3 Command User Role

11.3.4 Planner User Role

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (59)

11.3.5 Execution System Watch/Modify Role

The user may interact with the csc'ciitioii \y\toiir to \ \ ~ i ~ c l i t h e s ta te of execution and perhaps even to modify the behaviour of a \v01 I ( I 5 i t i i i i l ; i t i o t i i i r \\ Iiicli t hc esccutioii hystem i s operating.

11.3.6 System Developer Role

The system developer has access to t Ire. tliilgilost i ( iiitcrface of the systeni running within each agent. This is supported by the tliagiro\ti( i r i tc i facc of each 0-Plan2 agent. The hehaviour of this interface can be set and modific.tl I)! \( i t t iiig lo\ P I \ of diagnostics. iihiiig hiittons, ctc.

11.3.7 User Support to Coiltroller Role

11.3.8 User Support to Alternative h4nnager

11.3.9 User as System Builder

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (60)

12 Performance Issues and Instrumentation

0-Plan2 has been designed in such a \vay t h a t cotnpo~~eiits can be improved within the speci- fications adopted. Performance issues have 1 ) t ~ t i considered in establishing the interfaces aad protocols used. The current prototype often includes only very simple implementations of some of the components. The prototype is running in interpreted Coiiimoii Lisp a t present. However, extensive instrumentation and diagnohtic f‘ac.ilitics lia\.r been built into 0-Plan2 to allow for experimentation in future.

12.1 Architecture Performance

An early consideration for the O-Plati’L p i o j t c t M <i\ t o (\11>1ire tha t the agent orientated design would not introduce overlieads of cotiipittatioii \\ I i i ( 11 noultl 1w uuacceptal~le. ,A number of designs for the miilti-process structure reqiriiwl to \upport 0-Plan2 were discussed. These included shared memory processes and pro( ( I \ \ ( + i t IiicIi u \ ~ l a server for access t o the shared data elements. At the time that thew d i \ c i i \ \ i o i i c , \\ ( ~ r c tiiliilig place there n’ah little uniformity of handling concurrent processeh in (’oniiiioii l.i\p \ y h t ~ i i i s . Tests were conducted with complete 0-Plan2 systems which had only a t rii.ial I i ~ l o \ \ Ir1tlgcl \oiirce included. These ~vcre i~iiplemeiited in versions of Coininon Lisp and tli(I ( * latigiiago.

Two measures were tested:

Agent Latency This measure sho\vs t l i o ~ r i i t ~ i t r i i i i r i titire for an event a t the agent boundary to be noted by the Event Managor. cotiiiii~t t r i c a t ~ l to llie Agenda hIanager/C‘ontrolle~. trig- gered (where the trigger is 11 n l l ) , ~ ~ ~ i i t i ~ i l i i i c i ~ t c ~ d to a Iinowledge Source Platform (which is waiting and idle) and an a p p t ~ ~ p r i i ~ t (1 Iiiio\\lctlgc Sonrce activated on tlie platforill t o process the agenda en coi.i.c.\polltlillgjii(liii~ to t I I ( > o \ e t l t .

Agent Cycle Time This iiieasure slio\\ t l i (> i ~ i i i t i t i i ~ i i ~ i tiuie for a Iino~vledge Source to post an agenda entry back t o the Agenda lI~Illi1g(~t~/( ‘ontroller aiicl teriiiiiiate its processing, for the agenda entry to be triggered (I\ l i ( ~ i v 1 I](\ t t isget. i \ 111111 ). cominiiiiicated to a I<iiowledge Source Platform (which is Lvait iiig atrtl i t l l ( > ) a n d an appropriate Iinowledge Sonrce is activated on tlie platform to procw\ t I I V i~geii(Id t.iit1.y. l’liis corresponds to a single cycle of tlie agent internally wlien only otr(\ I\ iio\\ Ictlgc. 5 o u 1 ~ e Plat form is a \ ailahle.

Our main performance goal was to allon. 1 Iir gcncration of a plan with a few hundred nodes, which we judge would require 500-1000 iig(’litlil (y( lc4. i n about 3 minutes. Suhjectively, we judged that 3 minutes was a n acccl)t iI1)lt iwi.io(1 lot rl iihc’r to sit awaiting a result in our demonstrations. However, in the currc~tit i r i i ~ ) I ~ i i r c ~ i i t a t i o i i . soiiic’ tasks take coir~ideral~ly longer t 11 a 11 this.

I S

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (61)

12.2 Coiistraiiit Manager aiid Support, Routine Perforiiiance

Our experience with earlier AI planncr5 siic11 a s Noirliii and 0-Plan1 was that a large proportion of the time of a planner coiild be spent i i i pei.loriiiiiig hasic tasks on the plan network (such as deciding which nodes are ordered with rcspcct to others) ant1 in reasoning about how to satisfy or preserve conditions withiii the plan. Such fiiirctioiis have Been moclularised and provided as Coiistraiiit Managers (Graph Operat ions Procohhor. ‘I’inle Point Network Rfanager. TOME/GOST Manager, etc) and Support Roiitiiieh (Qi icst ion .I nsiwriiig, etc) in 0-Plan2 to allow for future improvements and replaceineilt by niorc cfficicii t vcrhions.

1 2.3 M o 11 it o rs and Iiis t r uiiieii t at i o 11

0-Plan2 incliides Rfoiiitoriiig and Inst riiiiic’iit <it ioii ~)acl\ageh to a h b i b t tlie tfeveloper and to allow 11s to address performance issues i n f l i t tire. I I I I)<irticiilar this nil1 allow 11s to identify tlie areas i n which processing time is being sl~citt for t l i f i r i c i i t s t ~ lcs of problem. It should allow us to confirm our assumptions on the propoi 1 ioii ol piwessiiig which takes place at the conqtraiiit inanagement and support routine 1e~c.l. 11.c ]in\ ( x oiilj . just reaclicd the stage where the 0-Plan2 system is complete enough to allo~v foi s i i ( 11 p ~ i loi I I I ~ I I C C iilstriiineniatioii to give us benefits. Monitoring aiid instrnmentatioii s t a t r i i t t b i i t s (‘[iii hc lac t t l at any point \v i th in the 0-Plan2 code and selectively enabled hy thc ( I C \ c~lop(~ i . The Monitoring package allows foi t l i i T ~ ~ ~ ~ r t lo\ P I \ of tliagnostics in the varioiih compone~its of 0-Plan2 and can be coiltrolled by tli(> I i i t c i f < i ( ( \ . \l~iidgci (‘oiitrol Panel t o etrsnre the developer receives the appropriate level of diagiiost i r s 101 tlio p i r t i c iilar task being u i ~ d ( ~ t a k ( ~ i r .

The Instnimentatioii package allow^ il iIiiiii1wi of ( o i i i r t iiig atitl clapbed or central processor time nieaE;urements t o be made. It alloit 5 for I ( W ~ I i i rig t I I C I \.[it.ious cou~iterh, for iiicreiiientiiig and dccrciiienting tlicm and for reatliiig out t I i ( ~ (.iiri(Iiit v d l i i ~ s . ‘l’lie 0-Plan2 prototype has heen instriimented in areas we consider sciihiti\c [i i it l f l i t i i i v orli \Till begin hystematic evaluation of the recordings taken by the iiistriiirirnls <is 0- I’laii2 i5 i’iiii on test prohlt~nis.

I!)

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (62)

13 Modularity, Interfaces and Protocols

This section provides a summary of coiit ri1)iitioii of‘ 1 1 i ~ O-I’Iaii2 project towards the identifica- tion of separable support iiiodiileh. i i i t c r i i a l i i ti0 C X I ( \ ~ Y I ~ I I interface specifications and protocols governing procwsing behaviours which arc> l ~ ( ~ l ( ~ \ ~ i 1 l i t 1 o ail A I plaiining system.

1 3.1 C omp on en t s

The 0-Plan2 project has sought to itlt.iit ify niocliilar coniponents within an A1 cominand, plan- ning a i d control system and to provide. c l c a ~ l y defined interfaces to these components and 111 o (111 les .

The main components are:

1.

2.

3.

4.

5.

13.2

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (63)

0 Monitors for oiitpiit inehhagcx. ctc .

il guideline for the provision of a good hiip1)oi.t i i iodii lc in ()-Plan2 is the ability to specify the calling requirements for the modnle i n a p w i x c \\ ;\y ( i .e. the .\,e t ~ s i f i v i f y rules under which the support module should be called I)), a liilo\~l(~tlg(~ soiirce or froin a coniponent of the architec- tiire).

13.3 Protocols

13.4 Iiiteriial Support Facilities

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (64)

5. n -

bC - - .- .A -

c 5 e - - -

- c CC bl:

d &

0 m

c 'e m

V c K

3 0 c

I I l l

$: E- c

*

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (65)

- <locking information> is provitlcd to allou itiforiiiatioii on whether this stage needs the plan state in REA11 niotlc 01’ l \ ~ l t l ’ ~ l * ~ tiiotle. If ]lot provided, the default as- sumption is that the stage i b a ltl<-Il) niotle stage and that all effects of the stage are created by cominiiiiicatioii with t h e controller (normally also saving information in the information field of t Iic iIg(ititla i~co r t l ivhen the knowledge source terminate5 if asked to do so at the stage c ~ n t l ) . I t is also possible to give information about the specific parts of the plan s t i l t ( ’ th i l t ca t i he READ by or \I‘RTTTEN to by this stage to allow for selective locliitig \triltt@txs to b e esplored in future versions of 0-Plan2.

e controller priority functioii. To provide ticuristic giiidaiice to the controller based npon the overall information in the agenda rccortl noiiiinating this kiiowletlgt source. This will only be applied to triggered agenda c w t r i w I t iiia). iiw Brw/)ch-f and Brunch-n infortiiatioii [IO] in a n agentla entry to pix ) \ i c l o I i c l i t i i \ i i c . giii(l<iiiw t o tlic c.oiittdlrr.

0 plan state poison handler. ‘1 Iir I’iiiic t i o t i I O I)(> c a l l t d \\ Itc~nclvti. / / l i s I i t i o ~ ~ l ~ d g e source terminates with a request to poiwi i ilia plriii \ l i l t ( ’ ( i . t ’ . \vlieti this litio\vlcdge source thiiiks that the plaik stat(. i5 iiicoii+tciil ‘ i i i t l t l i i i t il (iltitiot i w o v c ~ r froin the problem itself). This is not used n.itliiii t l i ( x ( III t ( 1 t i i \ ( i t \ i o i i o f O-Plati2.

13.4.2 Agenda ,Trigger Language

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (66)

13.4.3 Controller Priority Language

C’urrently, the 0-Plaii2 Controller wlcct 5 iigtlti(li1 rics h a v d on a 1111 tticrical priority wliicli is simply a statically computed iiicasiirv of t hr 1)riot ity of outstanding agenda entries in a plan state. Our aiin for the future is to provitlc a riilc I)iici(’(l colltroller which can make use of priority information provided in the form of riiIc5 i i i ail O-l’li>tl2 (“ontroller Priority Language. This concept will allow 11s t o clarify oiir idea, 011 Iiilt i nforiiiation hhoilltl govcrii controller ordering decisions. Domain information linking ro g(>ti(>ric (’oiit roller Priority T,angiiiige statements which can affect the controller decisioiir is lili(>ly to I) (> coiiciitlcretl ab part of‘ ii litlli hetwcn Task Formalism (TF) and the operation of t l i o ( ‘ot i t t~ollct~.

13,5 Ext eriial Interfaces

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (67)

14 Spacecraft Command and Control Application

0-Plan2 ha4 been tlemonstrated 011 ii i i i t tii1)oi of sti ir i l l al)plicatioiis during the development of the ideas and the prototype. The t i t lo of‘ t I i c ptojcct ”Spacecraft (’omiiiaiid and Control U4ng A I Plaiiiiiiig Techniques” - reflcctcd a ( howii dl)plication lo demonstrate the ideas being developed. The spacecraft plaiuiing and coiit I 01 tloiiiiiiii fomictl a useful example within which to consider the nccd to separate fuiictioiialitj. i t i tliffcieiit agcnts ivitli very different coinputatioii and I.eal-time response requirements.

This application shows the devclop~nciit of n pldii for tlic control of a siiiiple satellite we have called EUSAT (Edinburgh IJniversity Satc.llitc). ‘ I l i i b sntellite is based 011 the actual University of Surrey’s successfiil UOWT serie\ of satellites. Earlier research into the application of tash plaiiijing and scheduling at E d i n l ~ u ~ g l ~ l i r i \ i i i ( l i i t l c t l uorL 0 1 1 c-lcfining a Task Foriiialisiii de- scription for 0-Plan1 for a spacccral‘t s i i i i i l r i t to I 05 \’[ - 1 1 I)ut oinitting confidential informatioa (which we called ROGUSXT) [ ] G I . This \ \ a s 1 1 1 1 1 1101 cstctidcd i n the T - ~ C I I I C D scheduling system [I21 which took a scheduling per5peci i1.o a \ oi)pos(d t o a t ask plan~iiiig view as in 0-Plan1 and generated actual on-hoard coinputcr I ) i < i i j (oi i i t i i ~ i t i ~ l s . ‘ I ’ l i c ~ 0-l’lan2 project I:I~<.AT inotlel uses tlic saine spacecraft modcl as ROC;

.\ coinmniiicatioiis wiring liariicss cliagidiii foi I I \ \’I i \ h l i o \ \ i i i n Figure S.

1 1 .

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (68)

3. Horizon Sensor

4. Space Dust Analyser

5. Digital Voice Recording ( D I G I ‘ I ’ . I L I ~ I < R )

6. Charge Coupled Device ( C C D )

7. Particle Wave Experiment

The experiments are connected via a wries of su.itchr.5 to a tape recorder (DSR) and then to either a 70cm or 2m antenna for traiisiiiishion to I l i c gi~oiuid. Alternatively soiiie experinleiits can he connected directly to an anlciiiia lltrough l i n e 6 instead of passing through the D S R . One of the experiments, called thc Iligi l ’ a l l e i ~ . allon 5 lor a iiiehsage to be loaded into a tape recorder (the DCE) from the groiiiid i i l i ( l s i i l ) w 1 i i o i i t l > ~ rc-traiiwiitted a t a later time hack t o the ground. A s well as the series of P Y \ ) V ~ i i i i v i i l , . 1 IIP b o t v l l i t ( \ Iiiust alho sent1 telemetry data to the groiin d.

The movement of data from an espcriitirtit to at i i i i l t ( ’1 i l i i i i h iiiodelled as a >et of switch scttiiigs. Each switch has a valid set of i n p i t i 5 iiil(l oi i t l ) i i t 5 illltl i l icw at.(. describctl a:, fol1on.s:

spa ccc raft .

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (69)

The 0-Plan2 plaiiiiiiig agent ha\ h w i i (I(>iiioiist I c i t r d gcurrating a plan for sitcli a task and passing it to an 0-Plan2 arc1ritc.ctui.c Itnsc.ti VS(Y 11 t i o i i \J stc’rir for 5iniplc tlispatcli and nionitoriiig to take place.

Other related work a t Edinhiirgli has Ict l to t 1iv trio platiiiitrg h y s t ~ i n s for the European Space &p~cy. The first was the P l a n - E R ~ [10] t(11ir \v11ich could generate mission plans for the European Space Agency’s E R P - 1 spacecraft. ‘I‘ l i i \ I)rototype was built in the KEF, [23] knowledge rc-presentation system and iises a simple plan r c ] ) I ‘ r h e t i t atioii. A second system, OPTIMUM-AIV [3], is able to generate and support t h p c>sccritioii of ~)lpn\ for spacecraft assembly, iiitegratioii and verification. This second planner iise5 a (;oal Structure lmsecl plan representation workiiig alongside links to a traditional projrc.1 niauagc’iricti1 xup1mrt system (..ZRTE:?,IIS [%I).

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (70)

15 Related Projects

0-P lan2 is one of a set of projects a t IMinl~urgli grouped under the title of E L J R O P A (Edin- burgh University Research into Open Planning .\rcliitectures). The combined research of these projects cover issues in Iinowledge nasetl Plaiiiiing and Scheduling and are anchored around the two main, long term research projeclh of 0 - P l a n 2 and TOSC.4 (The Open SCheduling Ar- chitecture). TOSCA is a variant of the saiiic itlca\ applied to the area of operations management in the factory (job shop) environment [;I. TOC.(-.\ employs appropriate knowledge sources for its domain of application ( e . g . resoiirce ah\igii mcnt . 1,ottleiieck analysis) which operate on an emerging schedule state, similar to the notion of i l i c plan state mentioned above.

Another project is investigating teii11)oraI reprcscnt atioiis for Planning and Schediiliiig t o pro- vides a more flesiblr representation of plan\ r t ~ ~ ( l \ ( Iictlulcs I)a\cd on temporal logics. Plaiiiiiiig and Scheduling are often considercd to I ) c \ i n i i l , i i <i(ti\.itic\. though tlic reality i h that they are quite different. TIoivever there is i i i i t l o ~ ~ l ~ ~ a h l ~ <t g iwt tlcal of overlap. pariiciilarly with respect t o resource hantlling. Our aim is t o tlt\\x>lol) tlc\igii\ r t ~ i d architectures siiiictl to both types of problem and to develop as niuch coiiiiiioii gioi111(1 i \ po\\ihIe. 0 -Pla ti2 1)laJ.h a key role in this plan.

A student research project [31] is investigdtiiig t I I P ~ ~ ( ~ ( ~ i i i i ~ c i i ~ ~ ~ i i t ~ for a reacti1.e execution agent and exploring tlir 0 -Plan2 architectiiw to iiiwt 1 I I ( ~ I t>(luii(\iiieiit\.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (71)

16 Future Plans for 0-Plan2

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (72)

cln an in el s directly.

Work is already underway on a mort‘ coiiii)~(ili(~ii\iv(~ cxstv-ution systeiii I)ased on the 0-Plan2 architecture [31] which could replace I tic h i n i p l c O X C C I I I join system i n the current prototype as as drmonstration. This work is sctking to valitlafc o t i i i(1oah ahout the agent capahilitics iicetled for communication between a n eseciit iou

It is not envisaged that a great deal of ~vork nil1 bc carried out in the ncar fu tu re on tlie job assignment agent. ITowever. we have a dc\irc to iriiprove tlie qua l i ty of t l i r User Tinterface and tlir siipport available for tlic effective ivrit iiig of tloinaiu iiiformatioii a l m i t an application ( i n TF') , the specification and altwatioii of t i 1 5 l i h wt for the. planner and tsrcalion systeun, and the maiiittnance of a 11scr view of the \t i i t( ' of l ) l i l i i t l i I ~ ~ . osccutioii and t h e t>stcriial world inotlel.

tell1 ; I I I ( I ik 11lil1111er.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (73)

17 Concluding Remarks

0-Plan2 provides an Architcrturc in n-liicli different agents with coi~iniand (job assigniiient ), planning and execution monitoring roles can hc built. The architectnre seeks to bc1)iiriite out tlie following components:

0 the representation of the proressing capabilities of an agent ( i n A‘mu*/<d!/c .5’oirr

0 the computational facilities available to ptrforiii those capabilities ( t l i e pohsihly iiiiiltiple Knorl~ledge Source Plcitfornzs).

0 the Coizstrciint hloncigcrs and coiiinioiily used ,5‘uyyort Roufir2e.i which are i ibcf‘i i l i n tlie coiistrurtioii of command. plaiiiiiiig c i l i d (oiitrol \!-\tenis.

0 the decision inakiiig ahoiit \i-hitt the agent 5hould do nest (in the { ‘ o t ? / r o / / ( r ) . i i i i t l

0 the handling of communication l ) ( x t \ i w i i o i i o agent and others.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (74)

References

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (75)

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (76)

[33] Sacerdoti, E. A structure for planh i \ i i d l ) ( l i i i i \ ioiii-5. .-I r i i f i r inl I .Yorth Holland. 3.977.

[3 11 Sadeh, N. and Fox, M.S., PrFferrrnrc Prqxiguiion in Trii?poml/C'~ipririty Constraint Graphs, Computer Science Dept , C'arnegie-Nellon I'iii verhity, 19x8, Technical Report C'M TJ-CS-88- 193.

[ 421 'rate, A. & nrahble, B. 0-Plan'L: ('Iioiw Ortlc4irS \ I ~ c h a n i h m s in an AI Planning Architec- ture in Proceedings of the 1990 D.\l< l).\ \Yorksliop 0 1 1 Innovative Appi~oaches to Planning, Scheduling and Control, Sail IXc5go. ('aliforiiia. I'SX on 5-8 November 1990. published 1)y ~,~organ-Ic;aufi~~ann. .AIho I i l ldi\t( ' t l n i l 11 l~ . l ) ra l ) l> l (~ a \ .\TtlI-TR-M. '41.41, TJiiiversity of l<dinburgli.

*US. GOVERNMENT PRINTING OFFICE: 1992-710-093-60058

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (77)

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (78)

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (79)

MIS"

OF

ROME LABORATORY

Xome Laboratory plans and executes an interdisciplinary program in re- search, development, test, and technology transition in support of Air

for all Air Force platforms. I t also executes selected acquisition programs in several areas of expertise. Technical and engineering support within areas of competence is provided to ESD Program Offices (POs) and other

Rome Laboratory's technology supports other AFSC Product Divisions, the Air Force user community, and other D O D and non-DOD agencies. Rome Laboratory maintains tekhnical competence and research programs in areas including, but not limited to, communications, command and control, battle management, intelligence information processing, computational sciences and software producibility, wide area surveillance/sensors, signal proces- sing, solid state sciences, photonics, electromagnetic technology, mper- conductivity, and electronic reZiability/maintainability and testability.

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ...· SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (80)

SPACECRAFT COMMAND CONTROL USING AI PLANNING TECHNIQUES ... · SPACECRAFT COMMAND & CONTROL USING AI PLANNING TECHNIQUES - THE ... The O-Plan project began in 198 1. ... and other - [PDF Document] (2024)

FAQs

How is AI used in planning? ›

How does AI work in planning? As a tool, it's best at identifying patterns, extracting ideas, and providing insights that planners might overlook. By leveraging AI's machine learning, planners can make informed decisions by thoroughly understanding the data at hand and utilising well-structured information.

What is FSSP and BSSP in AI? ›

FSSP, also called progression planning, starts from the initial state and plans a sequence of actions to reach the goal state. BSSP, also called regression planning, starts from the goal state and works backward to determine which actions could have led to that state.

What is plan space planning in AI? ›

Plan-space planning uses a more general plan structure than a sequence of actions. In plan-space planning, planning is considered as two separate operations: (1) the choice of actions, and (2) the ordering of the chosen actions so as to achieve the goal.

What is the best AI for forecasting? ›

One of the top AI sales forecasting tools is Salesforce Einstein, which uses machine learning to analyze historical data and provide accurate predictions. Another popular tool is Clari, which uses AI to analyze sales data and provide real-time insights and forecasts.

Top Articles
Latest Posts
Recommended Articles
Article information

Author: Frankie Dare

Last Updated:

Views: 6371

Rating: 4.2 / 5 (53 voted)

Reviews: 92% of readers found this page helpful

Author information

Name: Frankie Dare

Birthday: 2000-01-27

Address: Suite 313 45115 Caridad Freeway, Port Barabaraville, MS 66713

Phone: +3769542039359

Job: Sales Manager

Hobby: Baton twirling, Stand-up comedy, Leather crafting, Rugby, tabletop games, Jigsaw puzzles, Air sports

Introduction: My name is Frankie Dare, I am a funny, beautiful, proud, fair, pleasant, cheerful, enthusiastic person who loves writing and wants to share my knowledge and understanding with you.