found atthe ENTCS Maro Home Page.
Flexible Communiation of Agents based on
FIPA-ACL
Jamshid Bagherzadeh 1
S. Arun-Kumar 2
Indian Institute of Tehnology Delhi
New Delhi, India
Abstrat
Communiation in multi-agent systems is an important subjet of the urrent re-
searh. In this paper, the syntax and semantis of a multi-agent programming
language, alled ECCS is dened. We fous speially on the ommuniation of
agents. The mainontributionof thispaperisa newand exiblewayof ommuni-
ationof agents. Wedesribea logitospeifydesirablepropertiesofagents. This
an be usedto verify the orretness of agents. We nally work out a well known
protoolas anexample.
Key words: Multi-agent systems, agent ommuniation
languages, CCS,FIPA-ACL.
1 Introdution
A multi-agent system is a olletion of agents, whih have their individual
goals (and perhaps someommongoals), and they ooperate with eahother
(through an agent ommuniation language), tofulll their goals. In the last
deadeseveralnewagentorientedlanguages(AOLs)andagentommuniation
languages (ACLs), like KQML and FIPA-ACL, have been dened. Two of
ACLshavegotmoreattentionintheliterature: KQML[3℄andFIPA-ACL[4℄.
The semantis of theseACLs havebeen dened interms of onditionson the
mental state of agents whih is supposed to have beliefs, desires, intentions
and so on.
We assume when an agent sends a message, it should satisfy some pre-
onditions, and moreover aftersendingthe message, itmightupdate itsloal
informationtosaveneessary informationwhihwe allthe eets ofsending
the message. On the other side, the reeiver might update its loal infor-
mation upon reeiving a message. In [1℄ we have dened a language, ECCS,
1
Email: jamshidse.iitd.ernet.in
2
Email: sakse.iitd.ernet.in
for multi-agent programming, and dened the semantis of the language in
terms of strutural operational semantis [10℄. One of the shortomings of
that semantis is the stati nature of preonditions and eets of a message,
i.e., the preonditions and eets of messages are xed and the programmer
annothangethem. Thishastwodrawbaks,rstly,aprogrammermightnot
likethe speiedonditions and eets hold and he mightliketodene some
other onditions and eets to be assumed during the sending or reeiving a
message. Seondly, the rationality of programs is low, in the sense that all
the agentsare treatedinthe same fashion, but the programmermight like to
have dierent agents ommuniating in dierent ways. For example sending
of a message to a trusted agent might have dierent eets than those of an
untrusted agent. In this paperwe have hanged ECCS toapture this issue.
To do this, we have assumed a programmer will dene a set of ommu-
niative ations whih we all performatives where eah performative has a
preondition and a post-ation. The preondition ould be a formula and
the post-ation is a proedure whih desribes the hanges in the informa-
tion store of the agent after ommuniation. For a partiular performative,
an agent eithersends or reeives the performative. Foreah of these irum-
stanes the orresponding preonditions and eets need to be dened. The
eetofsendingaperformativeisdirentfromtheeet ofreeivingthesame
performative. Any agent programhas itsown eets (perhaps dierent from
other agents).
The literature ontains several agent-oriented languages with a variety of
ommuniativeats(eg. AGENT-0[12℄,AgentSpeak(L)[11℄,GoGOLOG[7,5℄,
3APL [6℄, et.). Even the ats themselvesmay be dierent, they are allbeset
with the sameshortommings asECCS [1℄.
Although ommuniation issues have been onsidered inthese languages,
they formonly asmallsubset of CAs mostly from KQML(or FIPA-ACL).
Our attempt inthis paperis to dene a method whih gives enoughex-
ibility to the programmer to dene new performatives and dene their pre-
onditions and eets on both sides of the ommuniation. Thus the set of
ommuniation ats is not xed and the new ones an be added.
WeuseCCS[8℄andextenditwithsomeprimitiveCAsfromFIPA-ACL[4℄.
Wedeneastruturaloperationalsemantis[10℄ofthelanguage. Thispaperis
organizedasfollows. Insetion2weshowthestruture ofthe theinformation
store. Setion3denes the syntaxofagentprogramminglanguageECCSand
the semantis of ECCS is dened in setion 4. In setion 5 we work out the
FIPA-request protool, and setion 6 isthe onlusion.
2 Information Store
Let Ag =f1;:::;ng be a set of agents. Eah agent has an Information Store
(IS)whihinludesaset of beliefsandaset ofgoals. Weuse modaloperators
B
i
and G
i
tostand forbeliefsandgoals ofagenti2Ag respetively. Wehave
assumed a logi KD45 for belief operator and a logi KD for goals. These
axioms for belief operator are: (assumeF and H are twoformulas)
K: `B
i
(F )H))(B
i
F )B
i
H) 4: `B
i
F )B
i B
i F
D: `B
i
F ):B
i
:F 5: `:B
i
F )B
i :B
i F
In addition toabove axioms we dene following axioms whih represent rea-
sonable relationsbetween beliefs and goals of anagent.
B
i G
i
, G
i
B
i :G
i
, :G
i
2.1 TheLogiBG
n [2℄
Assume D is adomain representing the set of objets of the system, Const is
a set of onstants eah of them representing an objet of the domain D, Var
is a set of variablesymbols,and P =fp;q;r;:::gis a set of atomi prediates
suh that every atomi prediate has a predened arity. A term t is either
a onstant or a variable (variables are universally quantied over the largest
sope). Assume i2Ag is anagent, then formulas of BG
n
are dened as:
trueand false are formulas of BG
n
;
Anyatomiprediatep(t
1
;:::;t
n
)2P isinBG
n
,wheret
1
;:::;t
n
areterms;
IfF and H are inBG
n
then soare :F ; F _H ; F ^H ; B
i
F ; G
i F
We assume eah agent (say i) has a nite belief base, denoted
B
i , and
dened as:
Bi
f! j where ! is a losedBG
n
formulag
A formulaislosedif itdoesnot haveany freevariables. Notie thatbeliefor
goal bases may have formulas with nested modalities. The semantis of the
logi BG
n
is dened in afairly standard fashion in[2℄. Weassume the set of
beliefsis onsistent, i.e.,
B
i
6j=false.
Any agent i has a nite set of goals alled the goal base of i. The goal
base of agent i,denoted as
Gi
, is dened as:
Gi
f! j where! isa losed BG
n
formulag
Note that the goal base of an agent should be onsistent, and it may not
ontain twoinonsistent goals simultaneously. I.e.,
G
i
6j=false .
Thesatisfationrelationbetweenpairsofformh
i
;iwhere
i
=(
B
i
;
G
i )
is a (onsistent) informationstore of agenti, and 2BG
n
, is dened as:
i j=B
i i
B
i j=
i j=G
i
i 9! 2
G
i
; !j=
We dene satisfation relation for intention as:
i j=I
i i
i j=G
i and
i 6j=B
i .
First rulestates thatanagentbelievessomeformula, if isalogialonse-
quene of its belief base. The seond rule states that agent i has as a goal
if is a logial onsequene of some formula! 2
G
i
. The third rule states
that anagent intends aformula if is a onsequene of some goaland the
agent doesnot believe.
Inthispaperwedon'tdisussthebeliefandgoalrevisionindetail,asthey
have been disussed already in[2℄. We have alsogiven a sound and omplete
proof system in[2℄ forthe propositional fragment of BG
n
. We haveassumed
that the belief revision funtion dened there, works for both the belief and
goal bases. We assume funtion revise(
Bi
, ) updates the belief base with
, and revise(
G
i
, ) updates the goal base with . Moreover the operator
is used for deletingformulas fromthe belief or the goal bases. To delete a
formula from a belief base
Bi
, we perform revise(
Bi
, ). To remove
a formula from the goal base we use revise(
G
i
, ) whih removes the
goals whih imply.
3 Syntax of Agent Programming Language
The following BNFdenes the programming languageL.
::= 0 j ( ): j ( ): j update(O
i
!): j perform(Bat):
j query(!): j observe(l): j
1 +
2 j A(
!
t )
(1)
The intuitive meaning of dierent onstruts are as in CCS [8℄. Operators
( ) and ( ) are used for sending and reeivinginformation between agents
respetively,whereisanunidiretionalommuniation hannelbetween two
agents(with an input port and output port ) and is a message type or
performative suh as inform(!) or request(a). These performatives will be
explained in the sequel. Operators update(O
i
!) (where O
i
2 fB
i
;G
i g and
! 2 BG
n
) and query(!) are used respetively for updating and querying
the information store. observe(l) is used to observe the literal l from the
environment, where l = p or l = :p and p2 P is anatomi prediate. Bat
is an internal ation (basi ation) whih is dened by the programmer, and
perform(Bat) whould run Bat. Basi ationswould beexplained later.
The meaning of the omposition operators is as usual. a: is a proess
that an perform ation a and then beome the proess ,
1 +
2
is hoie,
whih means either
1 or
2
willbe exeuted. A(
!
t ) is the allof a label (or
proedure inimperative languages) withatual parameters
!
t .
Formally an ECCS agent is a tuple M
i
= h i, 0
B
i ,
0
G
i
, Bats, Ch, LB,
, P ats i where i is the identity of the agent, 0
Bi
is its initialbelief base
and 0
G
i
is initial goal base, Bats is a set of basi ations, and Ch is a list
of hannels whih are used by i forommuniation. LB isaset of proedures
whih we all labels,and 2 L is the main program. Finally P ats is a set
of performative ats tobeused in ommuniation.
Abasi ationisdened ashb(
!
x):(
!
x); (
!
x)i, whereb isthe nameof
the ation,
!
x is a set of formal parameters and pre(b(
!
x)) = (
!
x) 2 BG
n ,
and post(b(
!
x))= (
!
x)2BG
n
are itspre and postondition respetively. A
hannel is dened as (s,r) where is the name of hannel, s is the sender,
and r is the reeiver. Labels are dened as A(
!
x) =
def (
!
x), where 2 L
is a program of the grammar (1), and
!
x is a set of formal parameters. A
performative at(P at) issimilarto abasi ation,with a preondition,but
insteadofthe postondition,ithasapost-ationwhihisarestritedprogram
ofthegrammar(1)(withoutommuniationandlabels). AnyP atshows the
onditions tobesatisedbeforeommuniationandthe eetstobeahieved
after the ation. P ats would bedisussed ompletelyinthe setion 4.1.
Finally a multi-agent system is a parallel omposition of various agents
whih may ommuniatetogether. Its syntax isdened as:
M = M
1 j M
2
j :::jM
n
whereM
i
(i2Ag)are agents.
3.1 Performatives of FIPA-ACL
Some of the performatives of FIPA-ACL[4℄are presented in table 1.
perf. syntax semantis
inform hs;inform(r;!)i
FP:B
s
!^:B
s (Bif
r
!_Uif
r
!) RE:B
r
!
request hs;request(r;a)i
FP :(FP(a)[snr℄)^B
s
Agent(r;a)^:B
s I
r
Done(a)
RE:Done(a)
hs;agree(r;<s;at>;)i= hs;inform(r;)i where=I
s
Done(<s;at>;)
hs;refuse(r;<s;at>;)i= hs;disonrm(r;)i;hs;inform(r;)i
where=Feasible(<s;at>) , =^:Done(<s;at>)^:I
s
Done(<s;at>)
hs;failure(r;a;)i=hs;inform(r;)i
=(9e)Single(e)^Done(e;Feasible(a)^I
s
Done(a))^^:Done(a)^:I
s
Done(a)
Table 1
Some oftheperformativesof FIPA-ACL.
Thersttwoperformativesareprimitiveandthenext threeareomposite
performatives, whih are dened based onthe primitive ones. The syntax of
eah performative is of the form < s;at(r; ) >, where s is the sender, r is
the reeiver, at is the type of ation, and is the ontent of the message.
The semantis of eah performativeonsists of Feasibility Preondition(FP),
whih need to be satised before the at is performed, and Rational Eet
(RE), whih isthe eet expeted by the sender afterthe at isperformed.
In this table, U
r
! means that r is unertain about !. Although this op-
erator is dened in the semantis of FIPA-ACL performatives, its semantis
is vague, and we will ignore it in our framework. The problem of using the
unertainty operatorappearswhenweombineitwithbeliefsandgoalsinthe
ontext of proof method dened in [2℄. We believe that the proof methodin
this ase beomesvery diÆult.
Moreover, in this table, Bif
r
(!) = B
r
!_B
r
:!, Uif
r
(!) = U
r
! _U
r :!.
FP(a)[snr℄ denotes the part of the FPs of a whih are mental attitudes of
s, Agent(r,a) means that agent r an perform ation a, Done(a,) means
thatationaisdone(ifB
i
Done(a;))orintendedtobedone(ifI
i
Done(a;))
and was true just before the performane of the ation, Done(a)=Done(a,
true),Single(e)meansationeisanatomiationand Feasible(a)meansthat
preonditionsofationaaresatised. Wewillbaktotheseinthesetion4.1.
4 Semantis of ECCS
ThesemantisofECCSisdenedbyStruturalOperationalSemantis(SOS)[10℄.
Many of the semanti rules are extensions of those for CCS [8℄. Let P be a
set of atomi prediates, p 2 P be an atomi prediate, and l = p or l =:p
bea literal.
At
=fg[f( );( ) j is a ommuniative atg[fobserve(l)g
The ongurationofanagentinludesthosepartswhihmighthangeduring
exeution. For an agent, i, it is a tuple h
i
;
i
;
i
;did(i;a)i, where
i is the
ontinuationoftheagent'smainprogram,
i
isitsinformationstore,whihis
dened as
i
=(
B
i
;
G
i
)representing the beliefbase andthe goal baseof i
respetively,and
i
isavariable-valuebindingwhihbindsvariablestovalues.
did(i,a) speies that ationa was done by i in the previous transition.
Theoperationalsemantisofupdate, query andbasiations aredenedin
table2. It isassumed thatO2fB;Gg,! 2BG
n ,and
0
i
=( 0
B
i
; 0
G
i
)isthe
0
i
=revise(
i
;O
i
!
i )
hupdate(O
i
!):
i
;
i
;
i
;i!h
i
; 0
i
;
i
;did(i;)i
i j=!
i
hquery(!):
i
;
i
;
i
;i!h
i
;
i
;
i
;did(i;)i
i
j=pre(b(
!
t
i
=
!
x))and 0
i
=revise(
i
;post(b(
!
t
i
=
!
x)))
hperform(b(
!
t)):
i
;
i
;
i
;i!h
i
; 0
i
;
i
;did(i;)i
Table 2
Semantisofinternaloperations.
information store of agent i after revision. The operator query(!) is used to
query! fromthe informationstore andbind thefreevariables. Thisoperator
is exeuted if there is a ground substitution suh that !
i
is satised by
information store of agent i, otherwise it won't proeed. We assume is a
ground substitution suh that Free(!
i
)Dom(). Here Dom() isthe list
of variablesof and Free(!
i
)is the set of freevariablesof !
i .
In the semantis of basi ations, b(
!
x) is a basi ation, and
!
t
i is a
ground list of terms and (
!
t
i
=
!
x) is a substitution of the parameters of
!
x
with the orrespondingground terms from
!
t
i .
4.0.1 Observe:
observe(l(
!
t )) where l is a literal, is used for updating the informationof an
agent by observing some atomi prediate from the environment. There are
two purposes for observe: observing some new information fromthe environ-
ment or verifying the existing beliefsfrom the environment. In the following
we assume is a ground substitution suh that Free(
!
t
i
) Dom(), and
Env denotestheenvironmentanditontainsthetruthvalueofaset ofrelated
prediates.
3
9Envj=l(
!
t
i );
0
i
=revise(
i
;B
i l(
!
t
i ))
hobserve(l(
!
t )):
i
;
i
;
i
; i!h
i
; 0
i
;
i
;did(i;observe(l(
!
t
i )))i
Note that if l(
!
t
i
) is already believed, then the revision will not hange
the set of beliefs. If
!
t
i
is ground then the environment needs satisfy only
l(
!
t
i
). Finally if there is no substitution suh that l(
!
t
i
) is implied by
the environmentthen wehave the followingrule:
8:Env 6j=l(
!
t
i )
hobserve(l(
!
t)):
i
;
i
;
i
; i !h
i
; 0
i
;
i
;did(i;)i
In this rule if l(
!
t
i
) is ground then 0
Bi
= revise(
B
i
; B
i l(
!
t
i
)), i.e.
B
i l(
!
t
i
) willbe deleted from the set of beliefs. But if l(
!
t
i
) is not ground
then 0
Bi
=
B
i
. We assume at any time at most one substitution may be
observed for
!
t
i
. I.e., ifEnv has morethan one possiblesubstitution ,only
one of themwill beobserved.
4.0.2 Label:
LetA(
!
x)=
def (
!
x)beanagentname,and
!
t alistofterms. Thesemantis
of A(
!
t ) is dened as:
h(
!
t
i
=
!
x);
i
;
i
; i!h 0
; 0
i
; 0
i
;did(i;a)i
hA(
!
t );
i
;
i
; i!h 0
; 0
i
; 0
i
;did(i;a)i
3
Environment might be assumed as an agent, whih is ommuniated by the observe
ommand.
4.1 Communiation Operators
We have two operators for ommuniation of agents:
sr
(m) and
sr
(m) for
reeiving and sending informationm through hannel
sr 4
, respetively. We
assume m is of the form perf( ), where perf is a message type like inform or
request, and is the ontent of the message. Assume perf is a typial per-
formative. To send a message, the sender must satisfy some preonditions.
In addition there might be some revisions in the mental states of the agents
involved in the ommuniation. We don't x any onstant preondition and
revisions of mental state in the semantis of the ommuniation operators,
instead,welettheprogrammerdenethesepreonditionsand appropriatere-
visions for any partiular performative. The programmeris allowed todene
a performativeat (P at), perf
out
for any performative perf. This P at has
a preondition and a post-ation whih show the onditions to be satised
before sending the perf, and the revisions after sending. Similarly the pro-
grammer an dene a P at, perf
in
for any performative perf, to speify the
orresponding revisions afterreeiving a message.
Formallytheseationsaredenedas: perf
out
(r, )=h!(self;r; );(self;r; )i
(for sending), and perf
in
(s, ) = h ;(self;s; )i (for reeiving), where r,
are formalparameters, and !(self;r; ) isaformulaofBG
n
denotes the pre-
ondition of the performativeperf. (self;r; ) is aprogram of the following
grammar (2) whih shows the internal ations to be done by self immedi-
ately after the ommuniation. selfis the owner of the ation. Consideringa
proedure, instead of aformulainpost-ation,givesusmore exibilityto at
rationally. Usingthismethodanagentmaydodierentrevisionsaftersending
thesamemessagetodierentagents. Toavoidnonterminatingproedures,we
allowonly thesimple sequential programs withinternal ationswithoutom-
muniationats and reursion. The formalsyntax of this simpliedlanguage
is:
::= 0j update(O
i
!):j perform(Bat): jquery(!):j observe(p): j
1 +
2
(2)
Thelanguageobtainedfromthisgrammarisasubsetofthelanguageobtained
fromthegrammar(1). Weassumepre(perf
out
(r; ))=!andpost(perf
out
(r; ))=
. Notethat perf
in
does not haveany preondition usually.
Example 4.1 Assume wewanttodene aperformativealled informin our
framework. We might dene aP at inform
out
for agent i as:
inform
out
( r, ):
pre = B
self
^:B
self Bif
r
post = update( B
self B
r B
self ).
query( B
self
Trusts(r,self) ).
4
Rememberthatthehannel
sr
hasasendingport
sr
whihisusedbysandareeiving
port
sr
whih isusedbyr.
update( B
self B
r
^ G
self B
r )
Here self denotes the agent whih inludes the P at, i.e., agent i in this
example. Wemightdene a P at inform
in
for agenti:
inform
in
( s, )
post= update( B
self B
s ).
update( B
self G
s
Done(a)^ B
self B
s G
self
Done(a)).
query( Trustable(s) ). update( B
self )
where a 2 f
s(sel f)
(inform( )),
s(sel f)
(inform if( )) g
Preonditionsofinform
out
aresimilartothosegiven intable1. Initspost-
ation weaddB
sel f B
r B
sel f
, andif selfbelievesthat rtrusts self,then itwill
imply that r would believe and deletes the goal for B
r
beause self has
believed B
r
Intuitivelyselfafterreeivinginform( ) whihissentby s,believesthat s
believes ,beauseitsupposes thats issinere. Inadditionitwillimplythat
s doesnotintendanymore todoinform( ),and itdoesnotbelieveany more
that s believes that self intends a (self might already intend that s should
inform ). If self trusts s, then itbelieves too.
4.1.1 Output ation
Beforedeningthesemantisoftheoutputation,wedenesomerelatedon-
epts. Wedene Done(hi, i, )to desribe asituationwhere the expression
isdoneintheprevious stepby i,and 2BG
n
wastrue justbeforetheper-
formaneof. Weassumeisaprogramofthegrammar2. Done(<i;>;)
an be heked by looking at the previous ations of i whih are kept in the
onguration of i.
On the other hand an agent, say i, an not hek the previous ations of
another agent, say j. Thus Done(hj;i, ) should be stored expliitly in the
IS ofi. WhenDone(hi;i, )omes intheontextof agoaloperator,(likeG
i
Done(hi;i, ) orG
i
Done(hj;i, )), itis manipulatedas a prediate whih
should be expliitly kept in the information store. In general we have the
followingases:
B
i
Done(hi; i, ), Done(hi;i ,) /*if is reallydone by i */
B
i
Done(hj;i, )^ i6=j ,
i j=B
i
Done(hj;i, )
G
i
Done(hj;i, ) ,
i j=G
i
Done(hj;i, )
The next operator whihis used in this setionis Feasible() 5
, whih means
that is feasible, and its preonditions hold. A sequene of ations =
a:
0
is feasible in the onguration C if the rst ation, a, is feasible in the
ongurationC,andinthenextonguration,C 0
,obtainedfromtheeetsof
a on C,the sequene 0
isfeasible. This an be doneindutively by applying
5
Here mighthaveommuniationationalso.
theeets ofationsintheongurations. Ahoieofationsisfeasibleifone
of the hoies is feasible. If is a ommuniative at then its preonditions
is dened by the orresponding P at, otherwise if itis an internal (orbasi)
ation, then its preondition is dened in the orresponding basi ation. In
general we have the followingases:
i j=B
i
Feasible(a) i
i
j= pre(a) if a isan internalor ommuniativeation of i
Bi
j= Feasible(a) if a is an internalation of j 6=i
B
i j= B
j
pre(a) if a isa ommuniative at of j 6=i
i j=G
i
Feasible(a)i
G
i
j= pre(a) if a is aninternal oromuniativeat of i
G
i
j= pre(a) if a is aomuniativeat of j 6=i
It is not possible for i to make feasible an ation a, if a is an internal ation
of j 6=i, beause i doesnot knowthe preonditionsof internal ationsof j.
The next operator is Agent(i;a) whih denotes that i is an agent whih
an do the ation a. It is assumed that there is a global funtion Agt where
any agent i an read globally, but write loally, i.e. only the ations whih
an be done by i might be updated by i. This funtion is dened as Agt: Ag
At ! ftrue, falseg, whih given an agent andan ation,speies whether
the agent an perform the ation ornot. To hek B
i
Agent(j;a), we use the
funtion Agt (with parameters j and a). Cheking the nested formulas like
B
i B
k
Agent(j,a) alsois possible using the funtion Agt, and it willredue to
hekonlyAgt(j,a). Thisisbeause,iknows thatkhasaesstothisfuntion.
Finally let , 0
and 00
be proesses of grammar(2), , 0
, 00
be infor-
mation stores, , 0
, 00
be substitutions, and 2L be a main program. We
dene the big-step transitionrelation=) as:
h:; ;; i
=)h:; ;; i
h:; ;; i
!h 0
:;
0
; 0
; i ; h 0
:;
0
; 0
; i
=)h 00
:;
00
; 00
; i
h:; ;; i
=)h 00
:;
00
; 00
; i
The forth element of the ongurationis not important inthe transition=)
andweshowitby 0 0
. Nowwearereadytodenethe semantisoftheoutput
ation. Let i and j be the sender and the reeiver agents,
ij
be a hannel
from i to j, be the ontents of the message, !=pre(perf
out (j,
i
)), and =
post(perf
out (j,
i
)), Semantis of message sendingis dened as following:
i
j=!; h:
i
;
i
;
i
; i
)h
i
; 0
i
; 0
i
; i
h
ij
(perf( )):
i
;
i
;
i
; i !h
i
; 0
i
; 0
i
;did(i;
ij
(perf(
i )))i
Table 3representsthe default denitions of perf
out
(j; )) for someperfor-
mative. We have written the post-ations of the performatives as a simple
proedure, though a rational agent might have a omplex proedure. In the
table, we abbreviated self as s. Some disussion is needed about table 3.
request
out
( j, a) /*sdenotesself*/
pre= :B
s G
j
Done(hj;ai)^ B
s
Agent(j,a) ^I
s
Done(hj;ai)
post=revise(B
s B
j G
s
Done(hj;ai)). query(B
s
Trusts(j,s)). revise(B
s G
j
Done(hj;ai))
agree
out
( j, a)
pre=B
s ^:B
s Bif
j
post=revise(B
s B
j
), where=G
s
Done(hs;ai)
refuse
out
(j, a )
pre= B
s
:Feasible(a) ^B
s B
j
Agent(s,a)^ B
s
^ :B
s Bif
j
post=revise(B
s B
j
:Feasible(a)^B
s B
j
), where=:Done(hs;ai)^:G
s
Done(hs;ai)
failure
out
( j,a )
pre=B
s ^:B
s Bif
j
post=revise(B
s B
j )
=Done(hs; i,Feasible(a) ^G
s
Done(hs;ai))^ :Done(hs;ai)^:G
s
Done(hs;ai)
Table 3
Defaultdenitions ofperf
out (j, ).
In request
out
(j;a) the preondition is similar to those in the table 1 without
FP(a)[snj℄, and moreover self intends that a to be done. In its post-ation,
self will believe that j believes that self has a goal to do a and if self knows
that j trusts self, it will imply that j will reate a goal to do a. In ontrast
with table1,herewe donot onsider FP(a)[selfnj℄forthe sakeofsimpliity.
The preonditions of agree
out
are the same as those of inform
out
(dened
before)but!isreplaedwith ,whihisdenedas=G
sel f
Done(hself;ai)
and its post-ation simply adds a new formulaB
sel f B
j G
sel f
Done(hself;ai).
In the preondition of failure
out
, Done( ,Feasible(a) ^ G
sel f
Done(a)) means
that an ation has been done and before performane of that ation, a was
feasible and self had a goal to do a, although in both of the performatives
refuse
out
and failure
out
, self does not intend any more todo a (it is shown in
the denition of ).
4.1.2 Input ation
Assume perf
in
(i, )isaP at ofagentj, whereiisthe sender, =post( perf
in (
i,
j
)), and is a valuation of the free variables of
j
, whih will be
initialized during the ommuniation. This will be laried in the parallel
ompositionrule. The semantis of reeiving isdened as:
h:
j
;
j
;
j
; i
)h
j
; 0
j
; 0
j
; i
h
ij
(perf( )):
j
;
j
;
j
; i !h
j
; 0
j
; 0
j
;did(j;
ij
(perf(
j )))i
Table4representsthe defaultpost-ationofthe P at,perf
in
(i; )forvarious
performatives.
post(request
in
(i,a )) /*sdenotesself*/
revise(B
s G
i
Done(hs;ai)).
fquery(B
s
Feasible(a) ). revise(I
s
Done(hs;ai))
+query( B
s
: Feasible(a) ). revise(:G
s
Done(hs;ai))g
post(agree
in
( i,a ))
revise(B
s G
i
Done(hi;ai) )
post(refuse
in
( i,a))
revise(B
s
). revise(B
s B
i G
s
Done(hi;ai))
where=:Feasible(hi;ai)^ :Done(hi;ai)^:G
i
Done(hi;ai)
post(failure
in
( i,a ))
revise(B
s B
i ^B
s
). revise(B
s B
i G
s
Done(hi;ai))
=Done(hi; i, Feasible(a) ^G
i
Done(hi;ai))^:Done(hi;ai)^:G
i
Done(hi;ai).
Table 4
Default post-ation ofthe P atperf
in .
Let us explain the meaning of the P ats of the table 4. After reeiving
request(a),the reeiverself willometobelievethatiintendsatobedoneby
self anditbelievesthati believesthat selfis abletodoa. Ifaisfeasible then
self willintend to do a, otherwise if a is not feasible then self willnot intend
to doa. In post-ation of agree
in
(i,a), the reeiver self willbelievethat i has
intended to do a. In the refuse
in
(i,a), self believes that i believes and self
believes too. Moreoverself doesnot believe thati believes thatself intends
a to be done by i (see request
out
for understanding the reason for this). In
this post-ation states that a is not feasible and it is not done by i and it
is notintended anymore tobe done. The post-ation offailure
in
issimilar to
refuse
in
but with adierent denition of .
4.2 Summation and Parallel omposition
Semantis of summation
1 +
2
isdened asusual. Let a2At
then:
h
i
1
;
i
;
i
; i!h 0
i
1
; 0
i
; 0
i
;did(i;a)i
h
i1 +
i2
;
i
;
i
; i!h 0
i1
; 0
i
; 0
i
;did(i;a)i
h
i
2
;
i
;
i
; i!h 0
i
2
; 0
i
; 0
i
;did(i;a)i
h
i1 +
i2
;
i
;
i
; i!h 0
i2
; 0
i
; 0
i
;did(i;a)i
In the semantis of parallel omposition there are two ases:
1: Forany l 2fg[ literals:
h
i
;
i
;
i
; i!h 0
i
; 0
i
; 0
i
;did(i;l)i
[:::jh
i
;
i
;
i
; ij:::℄