ѧ
JOURNAL OF SOFTWARE
1999 10 7ڡNo.7 Vol.10 1999



agentϵͳһֽģ
㡡ޡ񴡡ʯһ
ժҪڶagentϵͳ(MAS),ͨŽagentʵЭҪ;.´ĽǶȶagentͨŽ̽з,agent˼ά״̬BDI(belief,desire,intention)ģͨŽ,һֽģ,֧ڻΪ֮ϵĶ͵Э̽,Խagent֪жȷĳͻ.оеıЭ̵ȷ,òģͿʵֻڳ,ʵ.
ؼʡagentϵͳ,,˼ά״̬,Э,Э.
ͼšTP18
An Interactive Strategy Model of Multi-agent System
LI Yi LUO Yi SHI Chun-yi
(Department of Computer Science and Technology Tsinghua University Beijing 100084)
Abstract Interactive communication is the main method to realize cooperation in multi-agent systems. In this paper, from the semantic perspective, the authors investigate the interactive communication procedure of multi-agent systems and incorporate the BDI (belief, desire, intention) model, the agent's mental state mode, into the interactive communication procedure. They present a model concerning the interactive strategy, which supports a variety of arranging interaction on basic interactive action to solve the conflicts of desires or intentions among agents. Compared with former other researches, such as negotiation through argumentation method, this strategy model can realize the flexible interactive communication based on different scenarios, and it is more practical.
Key wordsMulti-agent system, interactive strategy, mental state, cooperation, negotiation.
ڶagentϵͳ(MAS)AgentоɷΪͨŻƺЭ̷Ҫ,߼ȱϵ,ȱͨ.MASϵͳ,ҪʹagentܴӽлȡϢ,ͨŽ뷽ʽµ.,ͨŲӦЭıΪ,ͨĳ,һagentҪʹһagentĳ֪(belief)ĳͼ(intention)Ϊ.ÿͨžʲôش,ӦɽĿꡢһagent֪˽ӵеĸֶ֪εۺϾ1.ĽģͿĽ,ΪĿ(desire)ͨŷʽṩֶ.,ǰһĴ˫֪˽˫ͼ(intention)Թ滮,ԻȡԵŻĹ.ֻҪ֪ͼ͹滮Ĵѯ,ͿԱ൱Ľ,÷ΧֻagentΪﵽĿĶȡжĹ滮.Щ,һֶagentϵͳģ.
1 ͨŽõ
1.1 ͨŵ
һagentԻ仯Ԥʱ,ҪǵagentĻһǲԼ,Ԥ.Ϊ˸õԤ⻷Ǩǿжʵ,agentͨ.ͨagentıҪ,agentԵ.ͨŶҲһضĹ滮ж,agentĹԤе.,ͨŽagent˼ά״̬Ĵ.
agent໥ϵĽǶȿ,ͨŵУ
.Ԥ⣺agent aͨŵ֪agent b֪ƶagent bж,ǵagent a֪agent bͼ,Ըֱӵ˽agent bеĶ.
.ƣagent aĹ滮Ӧagent bɵĶͼagent b,agent b,жܵagent a滮Լ.agent aضϿagent b.
BDI,ͨŵУ
.֪(belief)֪ʶ,agentĹ۲췶Χ֪ʶֲagentĲ.
.(desire)˽agent,Ԥĳagentжagent,ԽЭ.
.ͼ(intention)˽agentͼ,Ԥĳagentжagentͼ,Խй滮Э(Э).
1.2 ͨŵ
ͨŵĲɱһĿ͹,ͨĴڵ.ͬʱ,ͨбнϢʽͼ.ͨýԻж,߶֪ԼͼᱻԷյ,һĹ滮ж.ͨŵĽշҲнϢ.
ͨҲagent֮䷢ȺΪ,ܽһagentϢ;ȫ.ͨŵķʵ,agentдԼѭͬͨŴʽ˼άʽagent֪.
ϢȡagentĹ۲Ϊ,滮ʶ2.agent aĳɵľƱһagent۲쵽,ݴ˽Эͬ,Ȼ֮ϢĴݲЭĽ,ڷжaȫԤ֮,Ĺ滮ûвκӰ.ڿa˼ά״̬Ǩʱ,˿ͨΪ,ֻ¼,ҪԹ滮.Ȥ,һֵ۲챻һ,ṩرϢ,ΪͨΪ,ǽ߲֪ⷢ͵ͨ.
ͨagent,ڹ滮ɺִйеͨΪӦȻ,ԼagentͼҪ,ͨѸжʹͼǰ¼.ϸ˵,ɷͷ,ÿ϶𸴺,¼žп϶,ڻÿ϶ǰ,ֻǼ.ͨЭ¼ֵ¿һƵ,agentĺԾ,һֻɼķжȷͼΪ϶ֵ,Ӵ˿ʼĹ滮һڹ۲쵽ԷĿ϶𸴺,ȷͼΪ϶ֵ,ʼĹ滮.
2 ͨŹ
ʵagentͨŵĹ£
(1) ͷsenderԼ˼뷭ͨԵĸʽ
(2) senderԸʽصͨŴ,ֺͼ
(3) 嵽շreceiver
(4) receiverȡеԴ룻
(5) receiver˼άռнԴ밴ʽΪ˼,ӶϤsender˼ά״̬.
ͨŹ漰󣺲ͨŵagentʹmedia.ΪȻԼ,һ뿪ͷ,Ͳȫɷͷ.ڱĵagentģ,ΪһagentĽṹ,agentԴ,ȫĴ,ܿڵշܵĴ.ڶ̬Żеͨ,Էͷ,ͨΪڶĴ.
agentͨžǸıϢ,巢͵agentĿɹ۲컷.ͨŹ۵չͨŵʽ,ͨ.,ĬʽЭ,Э˫ûֱӡȷӦ,˫ΪȻͨŵĶ.aж¶Ϊͼ,baͼнһƶϳabڴ.ֹ,ͨŽϢbԹ۲쵽¼,ͨŵЧbԴϢĽ.ֽǶȿ,ûͨŵЭҲͨŵ,ĲֱӵԤϢ,ڸжеϢں.agentĹ滮bԴϢĽ.ֽǶȿ,ûͨŵЭҲͨŵ,ĲֱӵϢ,ڸжеϢں.agentĹ滮ʶȻڴͨ.
2.1Ϣ
agent a֪пԽͨŵagent b,֪bЩԻΪģʽM,ҪݵϢI*.
(1) ͨCM(I*).Cһʵ,Ҳһж滮.
(2) Cʵ,䷢͵bĽջRcv(b).
(3) Cж滮,򰴹滮жִ.
(4) aĹ۲Ϣм¼ϢI*Ϊsent״̬.
2.2 Ϣ
Խb,ͨĹ۲һ.ͨŷͬ,Ҫ֪пͨŵĪagent aڡ֪aʹõԻжģʽ.շһջRcv(b),b˼ά״̬ͬص.ʱ,ͨڷշĻ֮,ջagentֱӰƵ(ͨ).ջص,ֱagentĽջȫص.ڼе߽ṹȫĽջ.һһϢͬһڵĻʵ϶Իȡ.ֹĽջй洢ĶUNIX.ȫĽջUNIXеIPCͨŵ䷽ʽ.
ɹ涨ֽջ,agentͨŽսṹ.
ջRcv(a){env-elmenv-elm״̬仯Աa۲쵽}
ȫֽջGRE(a){env-elmenv-elmRcv(a)b,env-elmRcv(b)}
ȫֽջagentĽջ.ȫֻеϢᱻagentյ.
Ⱥջ
TRE(a,TEAM){env-elmenv-elmRcv(a)bTEAM,env-elmRcv(b)}.
ȺջĳȺڲ,Ⱥ黷еϢᱻȺagentյ,һᱻȺagentյ.
Ⱥռջ
TPRE(a,TEAM){env-elmenv-elmRcv(a)bTEAM,env-elmRcv(b)TEAM,env-elmRcv(b)}.
ȺռջĳȺڲ,Ⱥ黷еϢᱻȺagentյ,ᱻȺagentյ.
ռջ
PRE(a){env-elmenv-elmRcv(a)b,env-elmRcv(b)}.
ռջagent˽н,ܱagent.
MAS,4ջļϾϵͳͨнղĽṹ.
磺
Rcv(a)GRE(a),
Rcv(b)GRE(b)TRE(b,{b,d})GRE(a)TRE(b,{b,d}),
Rcv(c)GRE(c)TRE(c,{c,d})GRE(a)TRE(c,{c,d}),
Rcv(d)GRE(d)TRE(d,{b,d})TRE(d,{c,d})PRE(d)GRE(a)TRE(d,{b,d})TRE(d,{c,d})PRE(d),
a,b,c,dĽշΧͼ1ʾ.dĽշΧ.

ͼ1agentջʾͼ
ͨŷͷ,ʹ䷢ͻշĽջص,ͨҲڻƶ.˿Լٶa֪bĴںbĳֽջԪʱ,Ϳ԰Ϣ͵bĻ.һʱ,ϵͳиagentagentĴڼջԪصϢ;ϵͳͨͼڷͲĽṹ,ջĹ˽ղĽṹ.ǹͬϵͳͨŽṹ,ϢԴݸЩagentĳagentЩϢͬʱЩagentյ.
ΪͨŸΪMASͨŻǽ,ҿָ֧֯.
չ̣֪aaģʽM,лRcv(b).
(1) ۲Rcv(b)ǷµĴC.
(2) CRcv(b),ת룺ѡʵģʽMϢI*=M-1(C).
(3) ڹ۲֪мSend(a,t,I*).
(4) ݶԷߵκTrust(b,a)ǷI*۲֪.
2.3 ͨŵĴ
ͨŵĴָͷʹշԼ֪ͼЧ.
ͨŹ,Ž׶,ͨŵֱӽǽ֪agent֪,Ⲣagentɴ֪.agentθagent֪ͼ޸Լ֪ͼ;agentΪʽ
abȫTRUST(a,b) iff BEL (a,t,BEL(b,t,prop))BEL(a,t,prop).
abȫOBEY(a,b) iff BEL (a,t,INT(b,t,prop))INT(a,t,prop).
һ,agentһı׼֪ͼ.
ͨŴɽմۺͽܴ۹.մȡͨʽضµŵͨ.Ϊͽմ,Խͨѹ.ܴȡͨ˫γ̶ԼշͨϢ̶,Ϣתеʧ.Ϊ˽ͽܴ,Բöͨʽ,߽շ̶ܵ.
3ͨ
agent˼ά״̬,ͨݿԷΪ֪.
3.1֪
֪ĴһagentһֻΪ,Ϊ߽ĳ֪͸һagent aʱ,ʵϾչ˽aĹ۲ͷΧ,aĻ֧.ĳЩ,Ƿͳ֪շ֪߼ͻʱ,֪ĴݾͳΪ˫۹̵һ(due process3),Ϊ˸ıһ֪,ıжͼ.֪ݶagentΪӰǼӵġԵ,ڶagentȺݻȴҪ.
3.2ͼ
ͼĴݶagentΪֱӵӰ,agentΪЭ.ͼagent˼ά״̬бΪе״̬,۲֪ͬı﷽ʽ,֪һؽд.ͼݵĳagent a,aӦǷΪԼͼֽй滮ж.
agentԭ,agentagentͼӦԼЧ.agentԺֱ׼ѡɽܵͼϻжѡ.
agentagentͼı׼,򵥵,DPS(distributed problem solving)Эagentķ.ɱΪAccept(a,desire-sent)TRUE.ڲ֯ṹ,agent֮仮ֳ,agentϼagentָ.ʱaccept(a,desire-sent)True,if Priority(sender(desire-sent))Priority(a).ںͬʽЭ,ÿӦıPay,agentǷһɱĵĲ.accept(a,desire-sent)True,iff Pay(desire-sent)Cost(Execute(desire-sent).жͼѡʱ,ѡȡPay-CostֵӦͼ.
3.3滮
һagent aĹ滮,Լٶagentɲж.a,ЩжaԤڵδ״̬,aͼﵽ״̬,ΪɵĿĹ滮ṩ.жһagentʱ䡢ռϵĳͻ޷,ǸȱΪ,ԼҪѱagentߵĴ,߽ԼԸеж.agentӦagentܡԸеĻ֪,ڴ֪ΧڼٶagentĻ.
֪agentĿܻĲУ
(1) ֻ֪й涨,ܼٶж.ַԺagent,ҪagentԻagent״н϶˽,ԱȽȶMAS
(2) ֪ȷ涨ɽеж,жԱٶ.ַҪagentԻagent̫˽,Ϊ,agent֮ΪȷʵΪν.
ֲԿԻʹ,ʹй涨ж,й涨ж޷ʱٲò(2).
agent֮乲ϢȫҼʱ,ô,һagent˽һagentĿΪƫ,Ӷһagent,ҲԸִеЭж,ʹ滮нϸߵĿִ,ٵĽתΪͼ,ʼִ.ϢҪķԴ,ҪϢۺЭ̴֮о,Լۺ㪩agentΪԵҪ.
ɹ滮,agent滮ͼ֣4ݣ
(1) ֻҪ滮ͼͼ,Ȩʽĺ
(2) Ҫagentͨ,ڻÿ϶𸴺żͼ,Эʽĺ.
4 ģ
涨ͨΪ
(1) ֪COMM(t,a,b,prop) abaprop.
(2) жREQ-COMM(t,a,b,action) abִaction涨Ķ.
ͨΪ,ɶagentĻӦ,ĿΪĶӦϵ.,SUREʾ϶֪,ASUMʾ֪,WILLʾȻ֪,POSSʾ֪.
4.1 Ӧ
һλab֪ϢabCOMM(t,a,b,mesg),bһͨĿܻӦΪ
(1) mesgSURE(a,t,prop), apropʱ baCOMM(t2,b,a,resp).
(1.1) bprop respSURE(b,t2,prop).
(1.2) bproprespSURE((b,t2,prop){optional:(propгͻ֪)SURE(b,t2,cont-prop)Infer(prop,cont-prop)FALSE}.
(2) mesgASUM(a,t,prop),apropʱ baCOMM(t2,b,a,resp).
(2.1) bprop: respASUM(b,t2,prop).
(2.2) bprop: respASUM(b,t2,prop){optional:(propгͻ֪)(SURE(b,t2,cont-prop)ASUM(b,t2,cont-prop))Infer(prop,cont-prop)FALSE}.
(3) mesgWILL(a,t,prop),aΪpropΪʱ baCOMM(t2,b,a,resp).
(3.1) bprop,t2Time(prop): respWILL(b,t2,prop).
(3.2) bprop,t2Time(prop): respSURE(b,t2,prop).
(3.3) bprop,t2Time(prop): respWILL(b,t2,prop){optional:(propгͻ֪)(WILL(b,t2,cont-prop)Infer(prop,cont-prop)FALSE}.
(3.4) bprop,t2Time(prop):respSURE(b,t2,prop){optional(propгͻ֪) (SURE(b,t2,cont-prop)Infer(prop,cont-prop)FALSE}.
(4) mesg=POSS(a,t,prop),aΪpropΪʱ baCOMM(t2,b,a,resp).
(4.1) bprop,t2Time(prop) respPOSS(b,t2,prop).
(4.2) bprop,t2Time(prop) respSURE(b,t2,prop).
(4.3) bprop,t2Time(prop) respWILL(b,t2,prop){optional:(propгͻ֪)(WILL(b,t2,cont-prop)Infer(prop,cont-prop)FALSE}.
(4.4) bprop,t2Time(prop) respSURE(b,t2,prop{optional:(propгͻ֪)(SURE(b,t2,cont-prop)Infer(prop,cont-prop)FALSE}.
͵ϢĻӦ,ƴ.
ھĽ,ϵͳݵʱĿִӦĶ.̲ǶԽյϢֱӷӦ,ҪʵԼͼ,ıagent֪ͼ.,ӦڻԵ½.
4.2
ضн,һϢĻӦҪݽĿġϴηϢԷӦϢ͹ڶԷ֪.
a𽻻յbӦ,bϢĻӦ.
(1) aĿģʹb֪prop.
(1.1) bĻӦ,˴ν.
(1.2) bĻӦܾ,Σ
(1.2.1) Ӧԭ,ҪشԭREQ-COMM(t,a,b,COMM(t1,b,a,reason)Infer(reason)BEL(t,b,prop));
(1.2.2) bӦԭreason,aжǷBEL(a,t,reason),
(1.2.2.1) BEL(a,t,reason),COMM(t,a,b,BEL(a,t,reason)option),
(1.2.2.2) BEL(a,t,reason),Infer(reason,ǰӦreason)prop,COMM(t,a,b,BEL(a,t,antireasInfer(antireas, reason)prop)),
(1.2.2.3) BEL(a,t,reason)(Infer(reason,ǰӦreason)prop),
.Ϊreasonϸڲ,Ҫ͸Ϣ:
REQ-COMM(t,a,b,COMM(t1,b,a,XXreason)),
.propreasonһԼ,еһ.
(2) aĿģһcondʹb׼ִaction.ΪЭ̹.
(2.1) bĻӦ,ȷǷҪbִaction.
(2.2) bĻӦܾ,Σ
(2.2.1) bӦԭ,Ҫشԭ REQ-COMM(t,a,b,COMM(t1,b,a,reason)Infer(reason)INT(t,b,action))
(2.2.2) bӦԭreason,aжǷBEL(a,t,reason),
(2.2.2.1) BEL(a,t,reason),COMM(t,a,b,BEL(a,t,reason)option),
(2.2.2.2)BEL(a,t,reason),(Infer(reason,ǰӦreason)INT(b,t,action)),COMM(t,a,b,BEL(a,t,antireason)(Infer(reason,ǰӦreason)INT(b,t,action))),
(2.2.2.3) BEL(a,t,reason)Infer(reason,ǰӦreason)INT(b,t,action),
.Ϊreasonϸڲ,Ҫ͸Ϣ
REQ-COMM(t,a,b,COMM(t1,b,a,XXreason)),
.жǷaction
.,COMM(t,a,b,Int(b,t1,action)).ʱӦINT(b,t,action)INT(b,t,worldInfer(action,world)FALSE),ں,ҪЭ,bworldͼ,ԼactionĹ滮
.,ѰҿactionҲreasonͻ,¼Ϊaction,REQ-COMM(t,a,b,action)
.Ѱµcond,ʹ(condreasonINT(b,t,action)),طREQ-COMM(t,a,b,actioncond).
5  
Ĵagent˼ά״̬ģ͵ĽǶȶagentͨŽá͹̽,һĻڳĽ.
1Ķ,ܻᵽĳ֪,Ϊ֪,Ϊ۲.֪ĲֻϵͳĿص,۲ĲɸݸԵĹ۲ֶΡǶȵۺϷ,Իȫȷ,ݴ˽Ե߼.
2ов,ΪΪҪԴ,Ҫ.ΪĳЩز߱,ʹԷִĳж,ҷЩϰԴʹԷִиж.
ϻԵۺ,ɸֽЭ.
ͨЭ̽ƵобЭ̣5,ɺĻ,֪ʶǿɱ.Эǿ,û.ĵĹǻ.
оõȻѧ.,1974,ʿ,ҪоΪֲʽ˹.,1971,ʿ,ʦ,ҪоΪֲʽ˹.ʯһ,1935,,ʿʦ,ҪоΪ˹.
ͨѶϵ:,100084,廪ѧѧ뼼ϵ
ߵλ廪ѧѧ뼼ϵ100084
ο
1Haddadi A, Sundermeyer K.Belief-Desire-Intention agent architectures.In:O'Hare G M P, Jennings N R eds. Foundations of distributed artificial intelligence. New York: John Wiley & Sons, Inc., 1996. 169185
2Rao A S, Murray G.Multi-agentmental-state recognition and its application to air-combat modelling. In: Lesser V R ed. Proceedings of the 13th International Workshop on Distributed Artificial Intelligence. Seatle, WA, 1994. 283304
3Hewitt C. Open infomation system semantics for distributed artificial intelligence. Artificial Intelligence, 1991,(47):79106
4Dunin-Keplicz B, Verbrugge R. Collective commitments. In: Proceedings of the 2nd International Conference on Multi-agent System. Tokyo: 1996. 5663
5Parsons S, Jennings N R. Negotiation through argumentationa preliminary report. In: Lesser V R ed. Proceedings of the International Conference on Multi-agent System. Tokyo, 1996. 267274
1998-03-05յԭ 
1998-07-20յ޸ĸ
