Some questions about interpreting plans as multi-stage influence diagrams

We have applied the SIPE system (a classical planning tool, see [1] ) to the Coast Guard’s problem of oil spill response. The Coast Guard has final responsibility for cleanup actions when an oil spill occurs in coastal waters. Oil spill actions fall into two types, "deploying" clean up equipment from the site near the coast where it is stored, and once deployment is complete, "employing" the equipment at the location of the spill. We call this the "oil spill tactical response" problem. Interestingly, for purposes of comparison in the literature, the tactical response problem has previously been formulated as a dynamic programming problem. [2]. Our intent is to incorporate some of the optimization features from a dynamic programming in the planner’s formulation of the problem. Our formulation of the domain encodes the Coast Guard’s operations as a set of planning operators that 1) choose equipment from a catalog of available equipment that is distributed by site, and 2) calculate the time at which the equipment would be deployed. By its nature, the problem is a race against the spread of the oil, so that cleanup actions, to be effective, must be completed within the time predicted for the oil’s landfall.