Towards a constraint-based multi-agent approach to complex applications

A society of agents based on constraint logic as a formalism that could support an agent environment for solving complex problems is explored. Within such society, the rights of individual agents and duties of these agents to others are often expressed in notions like communication, co-operation, negotiation, autonomy, and so on. The motivation comes from the fact that a contract net is essentially an imposition and cannot be easily reconciled with the notion of autonomy, social order, co-operation, and negotiation, and was related to a logistics problem. Our approach is to use constraint-based theories and methods to introduce an abstraction that can be used to articulate how the agents communicate, co-operate, negotiate, and how social laws are to be introduced. A CANET (Constraint-based multi-Agent Network) framework is proposed in which various societal notions can be put into operation. The CANET approach is based on relations, not on hierarchy. Within the CANET architecture, agents interact via a constraint store that consists of basic constraints; agents are treated as autonomous, reactive, pro-active, and as a social system. Each agent consists of layers of reactive, planning, and co-operation components. Communication between agents is treated as constraint passing. Co-operation and co-ordination are treated as constraint propagation whereas negotiation is treated as constraint relaxation. Social laws are treated as hard constraints. A prototype is developed that extends the scope of a logistics system that includes trucks operated by a number of companies. Fischer and Kuhn's (1993) approach has limitations in the sense that a contract net approach is presented, which is a highly regulated, ordered society and does not reflect preferences of agents. The role of constraint logic for such application is not explored. That fact holds in many other application areas. CANET is implemented in Oz (now called Mozart), an object-oriented language developed at DFKI (German Research Centre for Artificial Intelligence) for logistics application. A CANET agent is seen as `concurrent objects', and CANET multiagents as `concurrent objects with constraints'. We have extended the contract net message passing by constraint passing. That change facilitates communication by constraint passing. Further, we have shown that task allocation, co-operation, negotiation, social laws, and co-ordination can be discussed in a single framework based on constraint logic. We have developed a framework in which various societal notions can be operationalised. We have made a limited comparison with other approaches, notably with a contract net approach that allows us to look at the definition of an agent, agent architectures, agent communication language, and agency in different perspectives. This research encourages us to incorporate notions like learning and evolution within the CANET architecture, and apply these to complex applications. Acknowledgements I am very grateful to my supervisor Professor Khurshid Ahmad for invaluable guidance throughout every stage of this research and for his criticisms, comments and suggestions. I wish also to thank Professor Roland Price for his generous support and encouragement and to Dr David Pitt for his advice and feedback which helped me a lot to improve my ideas on my work. Many members of Al group, specially Andrew Salway and Steve Collingham have contributed greatly to my work during the years I have been working on this thesis. I would also like to acknowledge the contribution of Mrs Caroline Jones who offered helpful comments. Thanks are also due to Professor Gert Smolka, Professor Seif Heiridi, Christian Schultze, Martin Henz, Michael Mehl at DFKI for helpful discussions on Oz. I wish to express my sincere gratitude to Hydraulic Research plc, and the Department of Mathematical and Computing Sciences, University of Surrey. My warmest thanks go to my parents and my sister Kanjhana for their constant support and interest in my academic activities. My wife, Bama, has been a source of encouragement and advice. Many thanks for her love and support.