On Planning: What is to be Done?

Abstract

In this paper, we suggest research directions in domain-independent planning, addressing several open problems. In particular, we present: A modal truth criterion for the planning of overlapping actions with multiple agents. A "reduction" of planning in time to classical planning and a discussion of the relative power of the two methods. A formal framework towards describing planners with partial deductive closure (derived side effects). A discussion of failure analysis, error diagnosis, and recovery in domain-independent planning. An exploration of mathematical models of time (and space) proposed in naive physics. A discussion of the relative difficulty of classical planning, versus planning with a dense model of time.