John L. Pollock
Planning theory has traditionally made the assumption that the planner begins with all relevant knowledge for solving the problem. Planning agents operating in a complex and dynamic environment cannot make that assumption. They are simultaneously planning agents and epistemic agents, and the pursuit of knowledge is driven by the planning. In particular, the search for a plan initiates the search for threats. Because the epistemic investigation can be arbitrarily complex and may be non-terminating, this has the consequence that the planner cannot wait for the epistemic investigation to terminate before continuing the plan search. This in turn has the consequence that the planning cannot be done by a traditional algorithmic planner. It is argued that the planning must instead be done defeasibly, making the default assumption that there are no threats and then modifying plans as threats are discovered. This paper sketches how to build such a planner based upon the OSCAR defeasible reasoner. The resulting planner performs essentially the same search as UCPOP, but does it by reasoning defeasibly about plans rather than running conventional plan search algorithms. A beneficial side-effect of such defeasible planning is that updating plans in the face of changes in the agent’s beliefs (often reflecting changes in the world) can be done more efficiently than by replanning.