To present a sharp criticism of the approach known as universal planning, I begin by giving a precise definition of it. The key idea in this work is that an agent is working to achieve some goal and that to determine what to do next in the pursuit of this goal, the agent finds its current situation in a large table that prescribes the correct action to take. Of course, the action suggested by the table might simply be, "Think about your current situation and decide what to do next." This method is, in many ways, representative of the conventional approach to planning; however, what distinguishes universal plans from conventional plans is that the action suggested by a universal plan is always a primitive one that the agent can execute immediately (Agre and Chapman 1987; Drummond 1988; Kaelbling 1988; Nilsson 1989; Rosenschein and Kaelbling 1986; Schoppers 1987). Several authors have recently suggested that a possible approach to planning in uncertain domains is to analyze all possible situations beforehand and then store information about what to do in each.

This let Flakey drive along hallways with no dead reckoning or planning whatsoever. It seemed miraculous at the time; a situated automaton that knew things without needing any models. However, I thought of it as (sensor-driven) feedback control, versus (plan driven, eyes shut) feed-forward control. I then used Mike Georgeff's procedural reasoning system (PRS) to make Flakey not only drive but navigate an office building. In some respects this project succeeded: the robot's "domain knowledge" was nothing more than a static connection graph--no distances to drive, no widths of halls or doorways, no a priori obstacles--such information was acquired en route from sensory input.

Rather than begin by discussing the points where I seem to be in disagreement with Chapman and Schoppers, let me start with something about which we seem to concur: The work on reactive plans can be broken into two parts. First is the work on pure reactive plans, which specify actions for an agent to take in all situations. Second is the work on caching reactive plans, which specify actions in only some instances and are primarily used to store the results of previous planning activity. Although Chapman would object to the use of the word plan, the basic distinction is one that his reply appears to sanction. Strongly put, the argument in my initial article boiled down to two claims.

As the AI community has matured, the role of AI Magazine has continued to evolve. Rich outlines several ways that this community-wide publication can address the current needs of AI researchers, and encourages broad participation from community members.

Rather than begin by discussing the points where I seem to be in disagreement with Chapman and Schoppers, let me start with something about which we seem to concur: The work on reactive plans can be broken into two parts. First is the work on pure reac-tive plans, which specify actions for an agent to take in all situations. Second is the work on caching reactive plans, which specify actions in only some instances and are primarily used to store the results of previous planning activ-ity. Although Chapman would object to the use of the word plan, the basic distinction is one that his reply appears to sanction.

Several authors have recently suggested that a possible approach to planning in uncertain domains is to analyze all possible situations beforehand and then store information about what to do in each. The result is that a system can simply use its sensors to examine its domain and then decide what to do by finding its current situation in some sort of a table. The purpose of this article is to argue that even if the compile-time costs of the analysis are ignored, the size of the table must, in general, grow exponentially with the complexity of the domain. This growth makes it unlikely that this approach to planning will be able to deal with problems of an interesting size; one really needs the ability to do some amount of inference at run time. In other words, an effective approach to acting in uncertain domains cannot be to look and then leap; it must always be to look, to think, and only then to leap.