CLAIM : A Computational Language for Autonomous, Intelligent and Mobile agents

1
CLAIM A Computational Language for Autonomous,
Intelligent and Mobile agents

• Amal EL FALLAH-SEGHROUCHNI Alexandru SUNA

University of Paris 6
University of Paris 9
2
Motivations

• Think and implement in the same paradigm agent
• Meet the requirements of
• distribution
• migration

3
Talk Outline

• Programming distributed MAS
• Agent Oriented Languages
• Ambient Calculus
• CLAIM specifications
• Agents
• Agents Reasoning
• Example
• Implementation aspects
• SyMPA
• Mobility Protocol
• Mobility Completeness
• CLAIM Expressiveness
• Conclusion and perspectives

4
Agent Oriented Languages (1)
1993, Y.Shoham agent oriented programming
languages Agent-0, Agent-K, PLACA, AgentSpeak,
VIVA, 3APL, dMars
Agents characteristics

• mental state
• beliefs
• goals (AgentSpeak, VIVA, 3APL, dMars)
• obligations (Agent-0, Agent-K, PLACA)

reasoning mechanisms
capabilities/services
Agent-0, Agent-K, PLACA, AgentSpeak
5
Agent Oriented Languages (2)
Agents characteristics
communication

• agent reasoning, without communication
• 3APL, dMars
• message passing
• Agent-0, Agent-K, PLACA, AgentSpeak,VIVA

6
Ambient Calculus Cardelli and Gordon, 98

• Interest improve the modeling of agents'
  mobility

M capabilities x variable
n name in n enter into n
out n exit out of n open n open n
acid open from inside mv in n
move in n mv out n move out of n
P,Q processes MP ambient
(?n)P restriction 0 inactivity P
Q composition !P replication
M.P capability action (x).P input action
?M? output action
Reduction rules
7
Ambient Calculus Cardelli and Gordon, 98
enter action nin m.P Q mR ? mnP
Q R
exit action mnout m.P Q R ? nP
Q mR
replication !P ? !P P
open action open n.P nQ ? P Q
other rules mv in n.P nQ ? n P Q
nmv out n.P Q ? P nQ nacid.P Q
? P Q
Communication (x).P ?M? ? P x?M
8
Our CLAIM
An agent oriented programming language that
homogeneously combines
Intelligence, Autonomy and
Mobility Cognition, Interaction and
Concurrence
Agent Languages
Concurrent Languages

• goals
• knowledge
• capabilities
• reasoning

• communication
• primitives

• mobility
• primitives

9
Agent Description
defineAgent agentName parent null
agentName knowledge null knowledge
goals null goal messages null
message capabilities null
capability processes null process
agents null agentName
defineAgentClass className ... newAgent
agentName ( className )
10
Agents Components
hierarchical structure Ambient-Calculus like
knowledge agentName (capabilityS,message,effect
) proposition
goal proposition
11
Variables

• agentName
• capabilitySignature
• message
• goal
• effect

variable ?x , where x could be
Examples
forAllKnowledge(?Ag(cap1,?mess,?effect))
process(?Ag) ?Ag - all the agents in the
knowledge base that have the capacity
cap1 forAllAgents(?Ag className) process(?Ag)
?Ag - all the sub-agents of the current
agent that belong to the class
className
12
Agents Communication
send ( receiver, message , c, t )

• agentName
• ?Ag
• all (broadcast )
• this
• ?AgclassName (multicast)

• proposition
• messages concerning the knowledge
• tell ( knowledge )
• askAllCapabilities ( agentName, agentName )
• askIfCapabiliy ( agentName,agentName,capabilityS
  )
• achieveCapability ( agentName,agentName,capability
  S )
• removeCapability (agentName, capabilityS )
• mobility messages

13
Messages concerning knowledge
A.send(B,tell(knowledge))
? the knowledge is added in the knowledge base
A.send(B,askAllCapabilities(A,B))
? B.send(A,tell(B(capabilityS,message,effect )))
A.send(B,askIfCapabiliy(A,B,capabilityS))
? B.send(A,tell(B(capabilityS,message,effect )))
A.send(B,achieveCapabiliy(A,B,capabilityS))
? the capability is performed
A.send(B,removeCapability(A,capabilityS))
? the knowledge about this capability is removed
14
Mobility messages
openBy(Ag) - ask open permission openOK(Ag)
openNotOK(Ag) - the agent Ag accepts / doesnt
accept to be open wantOut(Ag) - ask exit
permission outOK(Ag) outNotOK(Ag) - the
agent Ag gives / doesnt give the
permission wantIn(Ag) - ask enter
permission inOK(Ag) inNotOK(Ag) - the agent
Ag gives / doesnt give the permission
15
Capabilities
capability capabilitySignature message
null message conditions null
condition do process effects
null proposition
condition function(args)
agentName.effect ( condition )
not condition condition and
condition condition or condition
16
Processes
P P.P proposition instruction funct
ionName(arguments) newAgent agentName(
agentClass ) in ( arg,agentName ) out (
arg,agentName ) moveTo( arg,agentName ) open (
agentName ) acid send ( receiver, message ,c
,t )
arg this clone process
17
Agents Reasoning
Forward reasoning

• choose a messages from the queue of messages
• find the capabilities that have this activation
  message
• verify the conditions of the chosen capabilities
• execute the corresponding processes

Backward reasoning

• choose a goal form the goal list
• find the capabilities that allow to achieve the
  goal
• verify the conditions of the chosen capabilities
• ? update the goals, ask for capabilities
• execute the corresponding processes

18
Example
e-commerce
Company agent

• knows the names of some Buyer agents
• can create Seller agents

Seller agent class

• can migrate
• can negotiate with Buyer agents

Buyer agent class

• can negotiate with Seller agents

forward reasoning
two solutions
backward reasoning
19
Agent definitions (forward - 1)
defineAgent Company parentnull 
knowledge B1(Buy,buy(),null)
B2(Buy,buy(),null)
goalsnull messagesnull 
capabilities sendSeller
messagesendSeller()  conditionnull do
newAgent S1(Seller).forAllKnowledge(?Ag(Buy,?mess?
eff)) send(S1,tell(?Ag(Buy,?mess,?eff))) .
send(S1,go())  effectsnull useData
messagearriveSeller(?Ag) conditionnull
do computeData() effectsnull
processes send(this,sendSeller())
agentsnull 
20
Agent definitions (forward - 2)
defineAgentClass Seller parentnull  knowledge
null goalsnull messagesnull  capabilities
goToBuyer messagego() conditionnull do
forAllKnowledge(?B(Buy,mess,?eff))
moveTo(this,?B).negotiate(?B)
.send(this,goToCompany())  effectsnull goToC
ompany message goToCompany() 
conditionnull do moveTo(this,Company).send(Co
mpany,arriveSeller(this)) effectsnull
processes null agentsnull 
21
Run steps (forward)
Company
S1 Seller
send(this,sendSeller)
22
Agent definitions (backward - 1)
defineAgent Company parentnull 
knowledge null goalshaveData()
messagesnull  capabilities useData
messagearriveSeller(?Ag) condition
?AgSeller.arriveToCompany() do
computeData() effectshaveData()
processes null agents S1 
23
Agent definitions (backward - 2)
defineAgentClass Seller parentCompany  knowled
ge B1(Buy,buy(),null) B2(Buy,buy(),null)
goalsnull messagesnull  capabilities goToB
uyer messagego() conditionnull do
forAllKnowledge(?B(Buy,mess,?eff))
moveTo(this,?B).negotiate(?B)
.send(this,goToCompany())  effectsendNegotiatio
n() goToCompany message goToCompany() 
conditionthis.endNegotiation() do
moveTo(this,Company).send(Company,arriveSeller(thi
s)) effectsarriveToCompany() processes 
null agentsnull 
24
Run steps (backward)
S1 Seller
Company
goalhaveData()
25
SyMPA Architecture(MASIF compliant)
Agent System
Agent System
Agent
Agent
Agent
Agent
I/O
I/O
I/O
I/O
I/O
I/O
I/O
Central System
26
Agent System Level
PSystem
27
Agent Level
28
Mobility Protocol
Central System
PSystem A
PSystem B
Agent A
Agent B
in(this,B)
29
Mobility Completeness
algorithm MobilityOperation Input agent,
SMA while (not SMA.structureCondition()) do
wait(t1) if (SMA.structureCondition())
then agent.saveState() agent.suspendAllProcesses()
agent.askPermission() while (not
agent.receivePermission() ) do wait(t2) if
(agent.receivePermission()) then SMA.update() wh
ile (not SMA.endUpdating()) do wait(t3) agent.res
toreState() agent.resumeAllProcesses() end
if else agent.removeProcess(this) agent.restoreS
tate() agent.resumeAllProcesses() end else end
if
30
Language Expressiveness
FIPA Contract Net Protocol
31
Agents Definitions (FIPA - 1)
defineAgent Initiator parentnull  knowledge
null goalsnull messagesnull  capabilities
calForProposals messagecallForProposa
ls() conditionnull do send(?AgParticipant,
cfp(this) )  effectsnull
rejectProposal message propsose(?PAg) condit
ion reject()  do send(?PAg,rejectProposal(thi
s)) effectsnull
32
Agents Definitions (FIPA - 2)
acceptProposal message propsose(?PAg)
condition accept()  do send(?PAg,acceptPro
posal(this)) effectsnull
processes send(this,callForProposals())
agentsnull 
33
Agents Definitions (FIPA - 3)
defineAgentClass Participant parentnull  kno
wledge null goalsnull messagesnull  capab
ilities refuzeCFP messagecfp(?IAg)
conditionnotAccept() do
send(?IAg,refuse(this) )  effectsnull
notUnderstood message cfp(?IAg)
conditionnotUnderstood() do
send(?IAg,notUnderstood(this) ) effectsnull
acceptCFP messagecfp(?IAg)
conditionaccept() do send(?IAg,propose(this
) )  effectsnull
34
Agents Definitions (FIPA - 4)
failure message acceptProposal(?IAg)
conditionfailure() do send(?IAg,failure(th
is) )  effectsnull done
message acceptProposal(?IAg) conditiondone()
do send(?IAg,informDone(this)
)  effectsnull ref message
acceptProposal(?IAg) conditionref() do
send(?IAg,informRef(this) )  effectsnull
processes null agentsnull 
35
Conclusion Main Contribution
CLAIM language

• Goals, Knowledge, capacities
• Backward reasoning reactive behavior
• Forward reasoning goal driven behavior
• Communication primitives
• Mobility primitives

Expressiveness
Completeness
SyMPA system

• Easy design of distributed MAS
• editing agents, interpret, agent interface
• Suitable platform for implementation
• management, creation, execution, authentication,
  migration of agents
• protocols for communication and mobility

36
Perspectives

• Define an operational semantics of CLAIM
• Enrich the language
• add security primitives and data types
• Extend de language possibilities
• develop libraries of agents
• endow agents with learning capacities
• Improve SYMPA
• call functions/subroutines defined in other
  languages
• offer several mechanisms for the management of
  agents and agent systems

37
References

• Luca Cardelli, "Abstractions for Mobile
  Computation", in Secure Internet Programming
  Security Issues for Mobile and Distributed
  Objects. Lecture Notes in Computer Science, Vol.
  1603, Springer, pp. 51-94, 1999.
• Luca Cardelli, "Mobile Ambients
  Synchronisation", SRC Technical Note, 1997.
• Luca Cardelli, A.D. Gordon, "Mobile Ambients",
  in Foundations of Software Science and
  Computational Structures, Maurice Nivat (Ed.),
  Lecture Notes in Computer Science, Vol. 1378,
  Springer, pp. 140-155, 1998
• Winton H E Davies, Peter Edwards, "Agent-K An
  Integration of AOP and KQLM", 1994.
• K.V.Hindriks, F.S.deBoer, W.van der Hoek,
  J.J.Ch.Meyer, "Agent Programming in 3APL", 1999
• M. d'Inverno, D. Kinny, M. Luck, M. Wooldridge,
  "A Formal Specification of dMARS", 1998.

38
References

• D. Milojicic, M. Breugst, I. Busse, J. Campbell,
  S. Covaci, B. Friedman, K. Kosaka, D. Lange, K.
  Ono, M. Oshima, C. Tham, S. Virdhagriswaran, J.
  White, "MASIF, The OMG Mobile Agent System
  Interoperability Facility", 1998.
• David Parks, "Agent-Oriented Programming A
  Practical Evaluation", 1997.
• Yoav Shoham, "Agent Oriented Programming",
  Artificial Intelligence (60), pp.51-92, 1993.
• Christian F. Tschudin, "Mobile Agent Security",
  in Intelligent Information Agents - Agent based
  information discovery and management on the
  Internet, Springer, pp. 431-445, 1999.
• Gerd Wagner, "VIVA Knowledge-Based Agent
  Programming", 1996.
• D.Weerasooriya, Anand S. Rao, K. Ramamohanarao,
  "Design of a Concurrent Agent-Oriented Language",
  1994.