1.06k likes | 1.19k Views
E 4 MAS 2005. 1. Responsibilities: What constitutes an agent in your model? What are its responsibilities? What is the environment in your model? What are its responsibilities? 2. Modeling: How do you model the environment?
E N D
1. Responsibilities: What constitutes an agent in your model? What are its responsibilities? What is the environment in your model? What are its responsibilities? • 2. Modeling: How do you model the environment? • 3*. Dependencies: Can agents be modified without modifying the environment? Can the environment be modified without modifying the agents? Session I: Definition, Scope, ModelsChair: Jim Odell
5' Holonic Modeling of Environments for Situated Multiagent Systems • 10' Overhearing and Direct Interactions: Point of View of an Active Environment, a Preliminary Study • 10' Environments for Situated Multiagent Systems: Beyond Infrastructure • 10' The Environment: an Essential Abstraction for Managing Complexity in MAS-based Manufacturing Control Session I: Definition, Scope, Models
Holonic Modelling of Environments forSituated Multi-Agent Systems Sebastian Rodriguez Vincent Hilaire Abder Koukam presented by Olivier Simonin Université de Technologie de Belfort-Montbéliard Systems and Transports Laboratory – Computer Science Team http://set.utbm.fr/info E4MAS’ 2005, July 26, Utrecht
E4MAS 2005 July 26th, 2005 Overhearing and Direct Interactions in MAS-Point of view of an Active Environment
Outline • Issues of overhearing in open MAS • Proposal: Understanding the Environment • Environment for Overhearing • Responsibilties & Features • Issues & Outlook • Summary
Overhearing • Overhearing • Indirect interaction type • Overhearer agent in Multi-Party Dialogue • Listen to a conversation • Known by conversation members • Not addressed by conversation members • Opportunity for `more interaction’ Conversation Overhearer
Issues with `current overhearing’ • Approaches • Multicast/Broadcast • Environment mediation • No application devoted to open system issues Environment-based overhearing is the `correct’ approach in Open Systems
Understanding the Environment • Environment as `correct’ abstraction • Deal with dynamicity and openness • Control and enforcement of overhearing • What kind of environment? • Responsibilities • Features • Control • Configuration
Environment Responsibilities • Purpose: Managing interactions • Overhearing (primary target) • Usual direct interaction (required target) • Methodology: Reflection from the environment
Environment Responsibilities • What E should process? Mediated Interaction • All interactions are executed by the environment • Overhearing becomes a function • Whom does E care of? Population and types • How does E execute processes? Environmental Rules • Where E should apply processes? Topology
Environment Model Features • Population • Agent Direct interaction & Overhearing • Elemental Direct interaction • Environmental Rules • Interaction modes • 2 modes in this work: `none’ and `overhearing’ • Properties that must be verified in the system • Enforcement by the environment
Environment Model Features • Topology • Communication Spaces • Logical and Physical (if necessary) • Location-aware ruling • Assignment of population and rules per CS
Evaluation • Purpose • Validate the model • Evaluate the cost of the environment process • Methodology • One scenario with repeated runs (with Jade) • No overhearing • Multicast-based overhearing • Environment-based overhearing • Overall result in short • Valid in the MAS scenario • Bottleneck with our environment implementation • 2500 agents up to 50% more time to completion
Issues and Outlook • Environment trade-off • Implementation bottleneck vs. modelling concept • Better implementation No bottleneck? • When? • No time perspective from the environment • Issues of propagation, action simultaneity • When is it necessary?
Summary • Overhearing is a promising interaction model • It requires an environment in open systems • Enact & rule overhearing • Natural framework • Systematic approach • `Natural’ distribution of responsibilities • Conceptually • Technically
5 September 2014 18 Environments for Situated Multiagent Systems: Beyond Infrastructure Danny Weyns, Giuseppe Vizzari, Tom Holvoet E4MAS, July 26th 2005, Utrecht
Modeling the Environment • Goal: help to clarify the confusion between • the concept environment • and the infrastructure on which the MAS is deployed • Approach: 3-layer MAS model • standard deployment model for distributed systems applied to MAS • agents and the environment first-order abstractions
Modeling the MAS • 3-layers • Top: MAS application layer • Middle: Software execution platform • Bottom: Physical Infrastructure • Agent, environment => crosscut the three layers!
Modeling the MAS Agent Environment Agent Agent Application Specific Logic MAS application layer MAS Framework Middleware Execution Platform layer Operating System Computer Hardware Physical Infrastructure layer Physical World
Responsibilities • We only consider MAS Application Layer • Agent • Autonomous entity • Act according to its design goal • Collectively solve a problem • Environment • >>
Responsibilities Environment • Domain-specific representation of problem context • Provides a space in which agents can perform their job • Provides a representation of resources to agents • Enabling entity • Enables agents to interact with domain resources • Enables agents to exchange information • Enables agents to coordinate behavior
Responsibilities Environment • Shields complexity to agents • Complexity of resource access • Complexity of interaction handling. • Complexity consistency management • Manages dynamics external to agents • E.g., digital pheromones, gradient fields, etc.
Conclusion • Environment and infrastructure are no synonyms • Agents and the environment • Both have an application specific representation • Both exploit and run on an execution platform • Both are part of the physical infrastructure • Environment is a powerful abstraction that can be used creatively in the design of a MAS solution
The Environment: an Essential Abstraction for Managing Complexity in MAS-based Manufacturing Control Paul Valckenaers & Tom Holvoet K.U.Leuven valckenaersp@acm.org
Responsibilities: Agents • PROSA • Product agents are 'recipe' experts • Resource agents manage factory resources • Order agents manage ongoing tasks • Staff agents give advice • Agents should not be functions • Ant agents • Exploring ants scout for solutions • Intention ants reserve capacity/slots on resources • Travel arrangement analogy
Responsibilities: Environment • Manufacturing resources • Factory • Machining station • Operator • Conveyor • Pallet • Decision-free aspects only • Pure reflection • Hotel analogy
Environment Modeling • Resource graph • Peer-to-peer: exits & entries • Conveyor belt feeding parts into a tunnel oven • Parent & child • Conveyor belt in a factory • Pallet on a conveyor belt • Lumped model • Position of a pallet on the conveyor belt • Supported operations/manufacturing processes/… • Attributes/methods/… handle non-graph aspects • Virtual navigation • Information spaces (stigmergy)
Dependencies • Modified agents without modifying the environment • Maintenance order/product agents • New product models • New resource mgt policies • … • Modified environment without modifying the agents • New resources • Removal of resources • New layout, changed connections • … • Incompleteness issue • New level of education • E.g. Communicating in probabilistic terms
1. Requirements: Does your approach target a particular domain? What kind of functionality does your approach offer? What kind of quality properties does your approach aim to realize? • 2. Design: What are the building blocks and their relationships to design the environment? What kind of support does your approach offer? • 3*. Agent-Environment: How do agents perceive the environment? How do they act in the environment? How do agents send and receive messages? Is the environment in your approach (in)dependent of the architecture of agents? Session II: Engineering EnvironmentsChair: Tom Holvoet
10' Engineering MAS Environment with Artifacts • 5' An Environment-Based Principle to Design Reactive Multiagent Systems for Problem Solving • 5' Landscape Abstractions for Agent-based Biodiversity Simulation • 10' An Architecture for MAS Simulation Environments Session II: Engineering Environments
Engineering MAS Environments with Artifacts Mirko Viroli, Andrea Omicini, Alessandro Ricci DEIS – Cesena Alma Mater Studiorum, Università di Bologna {mirko.viroli andrea.omicini a.ricci}@unibo.it
Outline We aim at developing a general framework for engineering MAS environments for cognitive agents • Motivation • Requirements • Agent-Environment • Design
Our motivation • Filling the “MAS-environment gap” • Standard MAS research (cognitive agents): • rooted on the intentional stance for agents • Studies on Environments (infrastructures): • providing services to black-box agents • We try to answer two fundamental questions • what is a fruitful way for agents to perceive the environment? • how to design a good environment for cognitive agents? • We develop on our previous work • Agent Coordination Contexts [Poster @ AAMAS2005] • Coordination Artifacts [AAMAS2004]
1. Requirements • Does your approach target a particular applicationdomain? • Not really, we aim to be general! Some scenario: • coordination, organisation infrastructures • workflow-based systems • What kind of (meta-)functionality does (will) your approach offer? • Abstractions & methodologies to the rational exploitation of environments • What kindof quality properties (non-functional requirements) does your approach aimto realize? • None, yet. Really an orthogonal aspect..
Our perspective • Environment as a set of tools or artifacts that agents exploit • to interact with sw & hw resources • either legacy or not • to participate to social activities • communication (e.g. message boxes, ...) • coordination (e.g. blackboards, schedulers,...) • inter-operation (e.g. dictionaries, yellow pages,...) • .... • in all cases: mediation • In other words, environment as a set of artifacts that agents use to achieve their individual as well as collective goals
A global picture • Any persistent entity is modelled and engineered either as an agent or an artifact • Artifacts play the role of the glue
2. Agent-Environment • How do agents perceive the environment? • as a set of artifacts • How do theyact in the environment? • by invoking operations provided by artifacts • How do agents send and receive messages? • messages?? more generally, interactions!!! • Is theenvironment in your approach (in)dependent of the architecture of agents? • Independent. But it supports/promotes agent rationality.
A Model of Artifact • Usage Interface • set of operations that can be executed by agents to use the artifact • Operating instructions • description of how the artifact is to be used to obtain its services • cognitive use by rational agents • Service provided • description of what kind of service is provided by the artifact • cognitive selection by rational agents • From formal aspects down to design/impl.
A picture interface agents an artifact
3. Design • What are the building blocks (and their relationships) todesign the environment? • At the first level: a flat space of artifacts • More in detail • artifacts can interact via linkability • an artifact can be something complex, e.g. itself a MAS with agents and artifacts • we are working on a library/taxonomy of artifacts • What kind of design support does your approachoffer to engineers? • Under study (see SODA methodology @ AOSE) • Mostly, it relies on the very difference between agents & artifacts
Example: tuple spaces out in rd ... Importing existing infrastructures
Example: pheromone infrastructure put pheromone perceive pheromone ... Importing existing environments
Example: schedulers, workflow engines get_task task_done ... Wrapping complex services
A First Taxonomy Some building blocks are already there • Boundary Artifacts • Coordination Artifacts • Resource Artifacts
Boundary Artifacts • Artifacts that model the interface towards the environment for an individual agent • Agent Coordination Contexts B B B
Coordination Artifacts • Artifacts that automatise a coordination task among a group of agents in the MAS • E.g. TuCSoN tuple centres C
R R Resource Artifacts • Artifacts that wrap resources of different kinds