This article includes a list of references, related reading, or external links, but its sources remain unclear because it lacks inline citations. (February 2011) |
Distributed artificial intelligence (DAI) is a subfield of artificial intelligence research dedicated to the development of distributed solutions for complex problems regarded as requiring intelligence. DAI is closely related to and a predecessor of the field of Multi-Agent Systems.
Objective
There are many reasons for wanting to distribute intelligence or cope with multi-agent systems. Mainstreams in DAI research include the following:
- Parallel problem solving: mainly deals with how classic artificial intelligence concepts can be modified, so that multiprocessor systems and clusters of computers can be used to speed up calculation.
- Distributed problem solving (DPS): the concept of agent, autonomous entities that can communicate with each other, was developed to serve as an abstraction for developing DPS systems. See below for further details.
- Multi-Agent Based Simulation (MABS): a branch of DAI that builds the foundation for simulations that need to analyze not only phenomena at macro level but also at micro level, as it is in many social simulation scenarios.
Agents and Multi-agent systems
Notion of Agents: Agents can be described as distinct entities with standard boundaries and interfaces designed for problem solving. Notion of Multi-Agents:Multi-Agent system is defined as a network of agents which are loosely coupled working as a single entity like society for problem solving that an individual agent cannot solve.
Software agents
The key concept used in DPS and MABS is the abstraction called software agents. An agent is a virtual (or physical) autonomous entity that has an understanding of its environment and acts upon it. An agent is usually able to communicate with other agents in the same system to achieve a common goal, that one agent alone could not achieve. This communicate system uses an agent communication language.
A first classification that is useful is to divide agents into:
- reactive agent – A reactive agent is not much more than an automaton that receives input, processes it and produces an output.
- deliberative agent – A deliberative agent in contrast should have an internal view of its environment and is able to follow its own plans.
- hybrid agent – A hybrid agent is a mixture of reactive and deliberative, that follows its own plans, but also sometimes directly reacts to external events without deliberation.
Well-recognized agent architectures that describe how an agent is internally structured are:
- Soar (a rule-based approach)
- BDI (Believe Desire Intention, a general architecture that describes how plans are made)
- InterRAP (A three-layer architecture, with a reactive, a deliberative and a social layer)
- PECS (Physics, Emotion, Cognition, Social, describes how those four parts influences the agents behavior).
Basic Disadvantages
The Basic Problems of Distributed AI are: 1.How to carry out communication and interaction of agents and which communication language or protocols should be used.
2.How to ensure the coherency of agents.
3.How to synthesise the results among 'intelligent agents' group by formulation,description,decomposition and allocation.
Real time Applications
- Problem Solving
- Multi-Agent Simulation
- Construction of Synthetic Worlds
- Collective Robotics
- Kenetic Program Design
Application in Industries
Demand for multi-agent systems
Industries have large number of different environments which are always interacting with each other.There are a number of intelligent agents which communicate with each other. Each agent has its own sphere of influence which is influenced or controlled by that agent.Spheres may be overlapping of some of the agents so there is a need to define dependency relationships between the agents.For achieving agile manufacturing conventional Computer integrated manufacturing techniques need to be modified.[2]
Demand of agility
Agile is a process of development in which continuous changes appear and process is successful in accommodating those changes. Manufacturers with shorter product cycle are able to keep track of needs in customer needs thus increasing market share.Agility requires system to be low in cost, having quick response, ability to correct itself in presence of faults. Agility has an impact on most part of product manufacture cycle and production system is one of them.
Characteristics of agility
| Cost | Response Time | Quality | Scope | Relation to Agility | |
|---|---|---|---|---|---|
| Agile | Low | Quick | Self-healing | Limitless | |
| Flexible | Low | Fast | Strong | Limited | Superseded |
| Lean | High | Slow | Fragile | Little | Complemented |
Problems with conventional CIM systems
Computer integrated manufacturing is a highly centralized system. It has a central computer which has all database and provides overall view of state of system for both internal representation and also interface to rest of environment.It also maintains schedule and is responsible for all activities dispatch, their scheduling and management.Problems with this approach are
- It depends on single system.
- Relies on global plans.
- Each action to be performed requires planning and scheduling.
These increase the problems of Manufacturing scheduling and control like
- Stochasticity- impact of external noise such as machine breakdown.[3]
- Tractability- impact of large number of interacting operations.
- Decidability- operational implications made from factory's turing machine equivalent.
- Formalchaos- sensitivity to initial conditions.
A centralised system is more susceptible to tractability because, the number of entities in industries interacting and to be managed is large and may lead to combinatorial explosion.Also it is difficult to implement agile systems in presence of such difficulties. Centralised systems are complex and costly.
Manufacturing scheduling and control(MSC) approaches
Heuristic Dispatching In Heuristic Dispatching a simple rule is used to decide the next step to be taken in process. It may be like select the process with shortest remaining process time. Many such rules are evaluated for their performance on parameters like total cost per job, number of late jobs, machine and labor utilization. Dispatching rules must be simple to implement. Only one heuristic may not be sufficient for one process to decide new step and a combination of many heuristics may be required.
Numerical Programming Linear and integer programming are often used to provide formulations of scheduling problems.New algorithms combined with powerful hardware make it possible to solve substantial problems using these techniques.They provide an approximate optimal solution. But they may get trapped in problem of combinatorial explosion.Also they are complex so large investment is required in sophisticated processors or have to agree to slower outputs.
Traditional AI It tries to implement satisficing approach to scheduling by accounting semantic information about ___domain, applying heuristic judiciously and selectively. Numerical solutions are not always possible due to cost and optimal solution solution is not always required.Symbolic computation has evolved which is faster than numeric but they are not suitable where configuration and load changes daily.
Advanced Heuristic It involves combination of all techniques depending on the system whose scheduling is to be done.Heuristic, simulation techniques and optimization principles are combined to give optimal solution. It is better than a single simple heuristic.
How can Multi agent system help?
Characteristics
Multi agent allows production system to be decentralised, each agent can takeits own action and allows concurrent execution rather than sequential.Autonomous agents have a local view of system replacing central computer with complete view of system and they have the ability to respond to an event locally. These agents perform depending on interactions in real time and their activities are not influenced by a global plan.Schedules emerge from independent decisions of local agents. There is no need of alternate cycles of scheduling and execution.Each agent is close to its own environment. Each action of agent is in sync with local environment and thus system adjusts automatically to the local noises and changes in agents.
When multi agent systems should be used?
Multi agent system is desirable in systems which require following characteristics in their systems
Distribution-Multi agent systems are distributed in nature as they are self dependent and take decisions based on their local environment. It is desirable to have distributed approach if a company has number of branches which need to work independently.
Factorisation-Power of industry production system comes from division of work among agents each responsible for that particular tool, entity and all information about it. This simplifies information gathering and operations.
Variability- It is easy to manage a system with conventional sequential scheduling if values of its variables do not change over time. But if values of variable vary large over time then multi agent must be used.
Taxonomy of multi agent system
There are several taxonomies introduced by multiagent system. Dimensions of Distributed artificial intelligence[4] are
- Agent granularity (coarse vs. fine)
- Heterogeneity of agent knowledge (redundant vs. specialized)
- Methods of distributing control (benevolent vs. competitive, team vs. hierarchical, static vs. shifting roles)
- and Communication possibilities (blackboard vs. messages, low-level vs. high-level, content)
In dimension 1 and 4 multiagent system have coarse granularity and high communication while in others they vary across whole range.
Taxonomy is also defined from application perspective.[5]
- System function
- Agent architecture(degree of heterogeneity, reactive vs. deliberative)
- System architecture(communication, protocols, human involvement)
Important contribution of this perspective is that it helps us differentiate between agent and system dimensions.
Various multiagent scenarios and issues arising in them are[6]
Homogeneous Non communicating agents
- reactive vs deliberative agents
- local or global perspective
- modeling of other agent state
- how to affect others
Heterogeneous Non communicating agent
- Benevolence vs competitiveness
- Evolving agents
- Modeling of others goal,action or knowledge
- Resource management
Heterogeneous communicating agent
- Understanding each other
- Planning communicative acts
- Benevolence vs competitiveness
- Resource management
- Commitment/decommitment
Three group of perspectives are important for multiagent system applied in industries.
Application function
In industries multiagent systems are used mainly for functions like production and design. Function of production includes scheduling to a large extent and to small extent monitoring and diagnosis.Such system endow distributed entities with some level of local intelligence.Systems which support design function have multiple agents which help designers coordinate their designs and thus reducing conflicts which helps in manufacturing.Some system also deal with manufacturing functions like power distribution, information integration and logistics.
Individual agent architecture
Here it seen how similar they are to each other to interact and how sophisticated is their reasoning.
How diverse are agents from each other? Agents in a multiagent system must interact with each other. For interaction there must be something in common with each other, they may be having different functions but they share common architecture which helps them in communication.Only digital messages is not only way to communicate but also through their effects on environment.in this case the common thing between agents is the sensor shared by both agents.
How sophisticated is individual agents reasoning? Tasks performed by agents vary from simple giving response to sensors to taking decisions similar to humans.So there is a need for agents to have a varying level of sophistication to suit the tasks assigned to them. Self consciousness level of sophistication requires that agents must know of the existence of other agents.Social level of sophistication can identify other agents state, goal and plans to modify its own action for better interaction.Artificial agents act as an interface to humans by providing them information and then getting commands from them thus they are augmenting human operator.
Combining reaction and planning [7] When we add more capabilities to individual agents by sophisticating their their reasoning capabilities we need to utilise it in a way as to benefit the system. This efficient utilisation can be achieved by combining reaction and planning .
Pure reaction-In this type there is no way to control the reaction. As there is no planning it is fast but inflexible.It is suitable in applications where all possibilities of tasks are known and there are no changes in it.
Reaction overridden by plan-In this planner is assigned to monitor the sensors and the reactions being taken by reactors on input from sensors.When reactor disagrees with the decision of planner ,planner pushes its own decision on the effectors.It is slow as decisions are planned by planner.This model is applicable in cases where there is uncertainty.Reactor takes care of anticipated requirements while planner takes care of uncertain situations.
Reaction modified by plan-In this planner can rewire the reactor in real time such that if a similar output comes from sensor, planner gives it to effector so that actions are performed at reaction speed.Rewiring is done by software.It is used in real time applications such as applying emergency corrections to real time systems.
See also
References
- ^ Ferber, Jacques. Multi-Agent System: An Introduction to Distributed Artificial Intelligence.
- ^ Wooldridge, Michael. MultiAgent Systems.
- ^ 5th international conference, CPAIOR 2008, Paris, France, May 20-23, 2008 : proceedings. Integration of AI and OR techniques in constraint programming for combinatorial optimization problems.
{{cite book}}: CS1 maint: multiple names: authors list (link) CS1 maint: numeric names: authors list (link) - ^ Keith S. Decker. Distributed problem solving: A survey.
- ^ N. R. Jennings (1996). Foundations of Distributed Artificial Intelligence. Wiley Interscience.
- ^ "http://www.cs.cmu.edu/afs/cs/usr/pstone/public/papers/97MAS-survey/node3.html#complexity".
{{cite web}}: External link in|title=|url=(help) - ^ "http://www.agent.ai/doc/upload/200302/paru96_2.pdf".
{{cite web}}: External link in|title=|url=(help)
Further reading
- Hewitt, Carl; and Jeff Inman (November/December 1991). "DAI Betwixt and Between: From 'Intelligent Agents' to Open Systems Science" IEEE Transactions on Systems, Man, and Cybernetics. Volume: 21 Issue: 6, pps. 1409–1419. ISSN 0018-9472
- Shoham, Yoav; Leyton-Brown, Kevin (2009). Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations. New York: Cambridge University Press. ISBN 978-0-521-89943-7.
- Sun, Ron, (2005). Cognition and Multi-Agent Interaction. New York: Cambridge University Press. ISBN 978-0521839648
- Vlassis (2008). A Concise Introduction to Multiagent Systems and Distributed Artificial Intelligence. San Rafael, CA: Morgan & Claypool Publishers. ISBN 978-1-598-29526-9.
{{cite book}}:|first2=missing|last2=(help)