LEARNING TO PLAN IN CONTINUOUS DOMAINS

In this paper, we propose an approach to planning in domains with cont inuous world features. We argue that current models of world change (i ncluding traditional planners, reactive systems, and many connectionis t systems) implicitly adopt a discrete action assumption which preclud es efficient reasoning about continuous world change. A formalism for continuous world change is outlined, and an ideal continuous domain pl anner is defined. An implemented computationally tractable approximati on to the ideal planner is discussed and its behavior is described. Em pirically, the implementation is shown to exhibit some of the importan t design features of the new planning approach. Learning plays a centr al role in this approach. With experience, accuracy is increased and p lanning time is reduced even though the system's background knowledge of the world is only approximate or ''plausible''. The acquired planni ng concepts are most accurate in situations similar to the ones in whi ch they are most exercised. Thus, the approach possesses a natural ada ptation to systematic properties implicit in the observed distribution of problems.