Finally, this algorithm is related to some of the hybrid systems that have been built for planning and decision making. Hendler (1988) describes a hybrid system in which a massive parallel component is used to provide heuristic information to a classical AI planner. A marker propagating network guides the classical planner towards more relevant plans. (Lehnert, 1987) describes a hybrid system that uses a stack and copy mechanism for control and numerical relaxation over a structured network for smooth decision making. The difference with the algorithm presented here is that in both these systems the control is still hierarchical and centralized, and might therefore turn out to be too inflexible for use in autonomous agents operating in a dynamic environment.