A recipe structure for plan execution in dynamic worlds

Abstract

In hostile and dynamic worlds, resource limited agents for problem solving will have to be both reactive and deliberative. However, for complex worlds, deliberative activities (which typically involve world modeling, goal generation, planning, negotiation, etc.) are not only time consuming, but also require greater expertise on the agent. There are several real world scenarios where an expert (a specialist agent) supplies the necessary knowledge to a resource-limited agent (a general agent) to solve its problem. We suggest that this knowledge be in the form of a recipe. We propose a recipe structure to encode the actions required to be executed through explicit planning, and the behaviors to be exhibited in time critical situations. The general agent translates the recipe into plans (as mental attributes) incrementally, and executes them. We also present the details of an implementation of this execution strategy, where we translate the recipe incrementally into a progressively abstract plan structure in a fire world.