Constraint-directed intelligent control in multi-agent problem solving

A framework is presented which demonstrates that much of the knowledge necessary for planning and coordinating cooperative activities among multiple problem-solving agents can be expressed in terms of constraints. The representation of a distributed problem-solving environment in terms of constraints yields many advantages. Many of these are at a low level and are provided by the constraint directed representation. However, the majority of the benefits of using constraints as a representation mechanism appear at high levels: using the organizations presented, it is possible to represent entire organizations and the knowledge within each agent in terms of constraints and relaxations. As evidence of his, a multiagent planning paradigm is presented that includes hierarchical nonlinear planning that utilizes constraints while planning task decomposition and distribution and monitors the execution of tasks. It is also shown that constraint relaxation provides a fruitful method of handling negotiations between agents when incompatibilities and conflicts arise