Multi-agent concepts and methodologies are finding increasing application in controlling complex, unpredictable systems in real time. Agent-like approaches routinely accept uncertainty and distribution, leading to control schemes where decision-making and responsibility are distributed much more widely than in conventional engineering practice. The successes that they have begun to produce are likely to lead to their heavier use, including use in applications that have time-critical aspects. In order for them to succeed in such situations, however, they will need development that addresses real-time and time-critical properties explicitly. This is the motivation for our SIG.
The subdivisions following the one immediately below are labelled in terms of applications, because at present we consider it desirable to drive the selection of basic-research topics by reference to concrete problems that are turning up in specific applications. This SIG will bring together researchers from a number of areas fitting into the paradigm, including:
- Adaptive real-time systems with knowledge-based components
- Manufacturing systems
- Robotics including multiple robot systems and robot football.
Adaptive real-time systems with knowledge based components includes traffic engineering applications. Multiple robot systems could include collections of robots which cooperate to perform some form of useful activity such as assembly operations or playing football or other games. The area of intelligent manufacturing systems aims to use multi-agent systems both as a modelling tool and a control software system.
The challenge for the SIG is to identify and focus on the general issues of coordination and control which are important for such applications, and to use the applications to stimulate and test advances in research on the issues, for example:
- How and where to represent information about the systems and the agent groups that operate on them
- Mechanisms (e.g. varieties of caching) for delivery of adequate responses to time-critical demands
- Coordination methods that are computationally economical enough for real-time use
- Decentralised management of limited common resources
- Adaptation of results from machine learning to collective learning in multi-agent systems
- New schemes of decentralised control to take account of the "real time" dimension
The "coordination and control" emphasis in this SIG is because the development of adequate methods for coordination and for control are crucial for the successful operation of multi-agent systems when a real-time discipline is imposed. Progress in other multi-agent topics is desirable too for real-time applications, but it will not be particularly helpful unless the problems of coordination and control are resolved adequately.
Current Activities
Second Meeting, Barcelona 6-7 July
Report on the First Meeting
A collection of PowerPoint slides from the first meeting, including the final plenary report by John Perram, are available on request. Just send an email to him (jperram@mip.sdu.dk) or to Bo Jørgensen (bnj@mip.sdu.dk). We will be establishing an ftp server when our systems people have recovered from getting the network to function in our new building.
