The use of supervenience in dynamic-world planning
Lee C. Spector, James Hendler · International Conference on Artificial Intelligence Planning Systems · 1994
This paper describes the use of supervenience in integrating planning and reaction in complex, dynamic environments. Supervenience is a form of abstraction with affinities both to abstraction in AI planning systems and to partitioning schemes in hierarchical control systems. The use of supervenience can be distilled to an easy-to-state constraint on the design of multilevel dynamic-world planning systems: world-knowledge up, goals down. We present the supervenience architecture which embodies this constraint, and contrast it to the subsumption architecture of Brooks. We describe the performance of an implementation of the supervenience architecture on a problem in the HomeBot domain, and we conclude with a discussion of the role that supervenience can play in future dynamic-world planning systems.