Designing Agents’ Behaviors and Interactions within ADELFE

Designing Agents’ Behaviors and Interactions within ADELFE
Gauthier Picard SMAC Team – IRIT, France Good afternoon, I’m Gauthier Picard. I’m coming from Toulouse France and working in the Cooperative MAS team of the IRIT. My talk will deal with Designing Agents’ Behaviors and Interactions in the framework of ADELFE.

ESAW'03 - Gauthier PICARD - http://www.irit.fr/SMAC
Contents Motivations and Research Positioning A Multi-Agent Theory An Agent Model Conclusion First, some concepts will be raised such as our view of self-organization and its use in the adaptive multi-agent system engineering. Second, I will expose our way to work with MAS paradigm. Then, I will present an Agent Model, the Cooperative Agent, that we use to design self-organizing societies. Finally I will conclude with some perspectives. 29/10/2003 ESAW'03 - Gauthier PICARD -

Some keywords Self-organization A system which changes its basic structure as a function of its experience and environment (Farley, Clark, 1954) Organism Level: The Morphogenesis of an Embryo Social Level: Ants, Termite, Flock Behavior, Democracy Emergence The global function of the system is emerging Its implementation does not depend on the knowledge of the collective function and must allow adaptation The self-organization of the components follows rules which are independent of the global function (georgé, 2003) The two main keywords we keep in mind when designing agents are Self-organization and emergence. A Self-organizing system is a system which changes its basic structure in terms of its environment and its own experience. Some good example are found in Biology or Ethologic. Emergence means the appearance of a new coherent function at the macro level. The implementation does not depend on the knowledge of the collective function  adaptation Emergence is possible by self-organization of the components that follow rules that are independent from the global function. 29/10/2003 ESAW'03 - Gauthier PICARD -

Why Self-Organization? Why Emergence?
Manageable systems  Organized by someone Self-organizing systems  Order comes from within and the function emerges from interactions It is more simple to design an agent than the global system when Complex: many components Open systems Dynamic environment No predefined global goal Not well specified… So, why focusing on self-organization or emergence? Typically, and in a near-past, engineers were faced to manageable systems where organization can be specified by someone. But, by now, engineers are faced to more complex problems and often need to use self-organizing systems where order and organization comes from within. Organization cannot be specified by someone and then has to emerge from self-organization. Moreover, it is more simple to design part of the system when this last one is complex, open, when the environment which it is in interaction with is dynamic, when there is no predefined global goal, and more generally when it is not well specified. Autonomic Computing, where the system has to maintain its own integrity and has to self-repair is a good example of too complex system to be specified. Ubiquitous Computing or Ambient Intelligence shows the difficulty to specify a global goal : how to satisfy the user who is plunged into an ubiquitous system? Human beings are a good example of dynamic environment too. Examples Autonomic Computing Ubiquitous Computing / Ambient Intelligence 29/10/2003 ESAW'03 - Gauthier PICARD -

Contents Motivations and Research Positioning A Multi-Agent Theory An Agent Model Conclusion So, now I will expound our view on designing self-organizing systems in which a coherent global function emerges from local interactions. 29/10/2003 ESAW'03 - Gauthier PICARD -

Adaptive Multi-Agent System Theory
Self-organizing multi-agent systems Engine of self-organization = agents’ cooperative attitude  cooperation failures detection and processing Emergence of a coherent global function Many applications : flood forecast, timetabling, on-line brokerage… ADELFE Method Process Notations Tools This theory deals with self-organizing systems where agents’ cooperative attitude is the engine of self-organization. This local criteria is a mean to ease exploring the search space. Cooperative attitude means that agents are able to detect local cooperation failures and to process them to reach back a cooperative state. As no global task is specified, only agents’ cooperative behaviors, the global coherent function of the system emerges from the organization of the MAS. Many applications in many domains were developped by using this principle such as flood forecast systems, timetabling problem or on-line brokerage. But, currently, our main topic of interest is to provide tools and notations to design such systems. This is why we have developped the ADELFE method, described by its process, notations and tools. 29/10/2003 ESAW'03 - Gauthier PICARD -

Adaptation The system has to compute the adequate function System adaptation MAS self-organization Agent Adaptation SYSTEM MAS AGENT  Beliefs/skills 29/10/2003 ESAW'03 - Gauthier PICARD -

Principle of Self-Organization
System Need for a criterion independent of the global function local to the agents Environment   Perception   Time t+1 : f*s + Action Time t : fs 29/10/2003 ESAW'03 - Gauthier PICARD -

Cooperation : Engine of Self-Organization
Cooperative attitude of an agent Definition of Cooperation All perceived signals must be understood without ambiguity (c1) The received information is useful for the agent’s reasoning (c2) Reasoning leads to useful actions towards other agents (c3) Proscriptive definition of cooperation Non Cooperative Situations (NCS) : ¬ c1 v ¬ c2 v ¬ c3 If an agent is in a NCS, it acts to reach a cooperative state Treatment : exception / cooperation failure 29/10/2003 ESAW'03 - Gauthier PICARD -

Functional Adequacy Theorem
Functionally Adequate Systems For any functionally adequate system in a given environment, there is a system having a cooperative internal medium which realizes an equivalent function Cooperative Systems Cooperative Internal Medium Systems 29/10/2003 ESAW'03 - Gauthier PICARD -

And So What? Functional Adequacy Theorem consequences To design a system, designers only have to ensure agents’ behaviors are cooperative to ensure the system provides a coherent function The behavior of the system will have to emerge from the cooperative interactions between agents  How can we design agents to obtain coherent societies? 29/10/2003 ESAW'03 - Gauthier PICARD -

Contents Motivations and Research Positioning A Multi-Agent Theory An Agent Model Cooperative agents’ modules Interaction aspects Fast prototyping Conclusion 29/10/2003 ESAW'03 - Gauthier PICARD -

Cooperative Agent Follows a three-step life-cycle perceive – decide – act Is composed of different modules Is cooperative (avoiding NCSs) 29/10/2003 ESAW'03 - Gauthier PICARD -

Cooperative Agent Interaction Perception Actions Sensor Actuator Representations Aptitudes Sensor Actuator Sensor Actuator Sensor Actuator Skills Sensor Actuator Sensor Actuator Sensor Cooperation Actuator Sensor Actuator Sensor Actuator 29/10/2003 ESAW'03 - Gauthier PICARD -

Interaction Module Interface between the agent and its environment Physical  sensors, actuators Social  mailbox and mail services Sub-modules Perception Action Interaction languages to communicate with other agents Triggered during perception and action phases 29/10/2003 ESAW'03 - Gauthier PICARD -

Representation Module
Partial knowledge on the environment the other agents itself Example of implementation Classical or fuzzy knowledge base Multi-agent system (like semantic organization) Learning capabilities Triggered during the decision phase 29/10/2003 ESAW'03 - Gauthier PICARD -

Skill Module Common knowledge on a particular field Example of implementation Classical or fuzzy knowledge base Multi-agent system (like semantic organization) Learning capabilities Triggered during the decision phase 29/10/2003 ESAW'03 - Gauthier PICARD -

Aptitude Module Capabilities to reason on representations skills perceptions Example : inference engine Returns an action to do Triggered during the decision phase 29/10/2003 ESAW'03 - Gauthier PICARD -

Cooperation Module Represents the agent’s cooperative attitude Rules set Conditions in terms of representations, perceptions, skills Actions : actions, representation modification Chosen cooperative actions are top priority Triggered during the decision phase 29/10/2003 ESAW'03 - Gauthier PICARD -

Functioning Perception Actions Sensor Actuators Representations Aptitudes Sensor Actuators Sensor Actuators Sensor Actuators Skills Stimuli Sensor Actuators Actions Sensor Actuators Sensor Actuators Cooperation Sensor Actuators Sensor Actuators Interaction 29/10/2003 ESAW'03 - Gauthier PICARD -

How to use this model? (an example)
Use of UML Definition of “cooperative agent” specific stereotypes Coherency rules to ensure the right use of modules 29/10/2003 ESAW'03 - Gauthier PICARD -

Interaction aspects A-UML interaction diagrams used to specify interaction languages and protocols As cooperative interactions lead the system to the right function : NCS treatment at the interaction level is needed 29/10/2003 ESAW'03 - Gauthier PICARD -

NCS Rules Definition Problem
Defining problems at the local level is easier than defining them at the global level BUT how can we ensure the NCS rule set is exhaustive? No formal solution, only hints and guidances Following the ADELFE methodology and AMAS definitions Fast prototyping 29/10/2003 ESAW'03 - Gauthier PICARD -

Fast Prototyping By using appropriate designing and simulation tool (e.g. OpenTool) By simply defining agents’ behaviors Pre-code the modules Dynamical aspects (state machines or protocols) By simulating behaviors Especially state machines Transforming non state machine models into state machines automatically Transforming protocol diagrams into state machines 29/10/2003 ESAW'03 - Gauthier PICARD -

Transformation into Finite State Machines
Procedure For each protocol, a concurrent state machine is associated A state is associated with each role involved in the protocol A sub-state of the role state is associated with each message reception and emission A transition between two states is created when a message is received or sent A-UML specific branches an aptitude is associated with the decision-making process of a OR or a XOR branch If the XOR node has n branches, it generates n transitions A state is created for each AND nodes 29/10/2003 ESAW'03 - Gauthier PICARD -

Example: Buyer/Seller Protocol
29/10/2003 ESAW'03 - Gauthier PICARD -

Seller’s Concurrent Machines

Seller Role Sub-State Machine

Contents Motivations and Research Positioning A Multi-Agent Theory An Agent Model Conclusion 29/10/2003 ESAW'03 - Gauthier PICARD -

Conclusion To design more adaptive systems Multi-agent systems Self-organization led by cooperation The function emerges from the interactions within a society of cooperative agents To design such systems, we propose the ADELFE Method ( Process (inspired from RUP) Notations (UML / A-UML) Tools (e.g. OpenTool, AdelfeToolkit, AMAS Adequacy…) To design agents of such systems, ADELFE proposes UML stereotypes AIP protocols Simulation of dynamical behaviors 29/10/2003 ESAW'03 - Gauthier PICARD -

THANK YOU 29/10/2003 ESAW'03 - Gauthier PICARD -

Designing Agents’ Behaviors and Interactions within ADELFE

Similar presentations

Presentation on theme: "Designing Agents’ Behaviors and Interactions within ADELFE"— Presentation transcript:

Similar presentations

About project

Feedback

Log in

Auth with social network:

Designing Agents’ Behaviors and Interactions within ADELFE

Similar presentations

Presentation on theme: "Designing Agents’ Behaviors and Interactions within ADELFE"— Presentation transcript:

Similar presentations

About project

Feedback