School of ECM University of Surrey Guildford, Surrey GU2 5XH, UK

Tel: +44 (0)1483 259823 Fax: +44 (0)1483 876051

The considered domains tend to be large, sophisticated, unpredictable, and have a number of modular components. In addition, there are various knowledge sources that must be consulted.

Research in distributed artificial intelligence (DAI) is concerned with solving problems in these domains with the help of knowledge sources that have a degree of autonomy, recognise when to apply the knowledge, and are able to co-ordinate such application with other knowledge sources. These autonomous 'socially aware' entities are called agents. A multi-agent system is concerned with co-ordinating intelligent behaviour among a collection of autonomous intelligent agents and co-ordinating their knowledge, goals, skills, and plans jointly to take action to solve problems.

We are investigating a constraint-based framework for simulating interactions between autonomous knowledge sources. Our hypothesis is that an agent's behaviour may be constrained and the interaction between agents may be related to constraint-based mechanisms. Within a society of agents, the agents interact by means of co-operation, negotiation, task allocation, and may obey social laws. Because a constraint can be a number, a message, a method, partial information and so on, and it has also been shown that a variety of computations can be performed by simply passing constraints, the constraint-based interaction approach is well-suited than the traditional mechanisms. In addition, this constraint-based communication approach can also be integrated within a contract net approach. Moreover, agent's goals may well interact by constraint-based approach.

We have been considering the simulation of a logistics system in which a number of trucks are expected to deliver goods to a number of cities, efficiently and cheaply. The system is implemented in Oz language (a concurrent constraint language developed by DFKI, German Research Centre for Artificial Intelligence). In this system, input from diverse sources is required and a number of processing algorithms are used. Such a simulation can be understood intuitively and now computationally in a multi-agent environment. We have shown that agents may communicate constraints between themselves within a contract net approach. The front-end of that logistics simulation is shown in the next figure. This work relates the SAFE-DIS project particularly with respect to how we may use an agent-facilitated simulation for the safe design of complex artefacts.