The techniques, described in this paper, are aimed at helping to solve the problems that arise in large unique logistic problems. Examples of these problems are: Organizing the medical teams and support equipment to respond to a natural/man-made disaster, Organizing the teams and equipment to fight a forest fire, and Organizing the troops and equipment to accomplish some military goal. The characteristic issues of these problems are:
One way to handle this domain, is to plan out the entire domain using the resources in the planning task. However, given the number of possible goals (thousands), and since all the goals interact over resources, this complete planning approach is computationally infeasible.
Instead this paper proposes to create a set of new plans (or appropriately modify old plans) and then check these plans for feasibility by scheduling the plans to use the available resources. By itself, this is not a new idea. However, we plan on interleaving the planning and scheduling so that planning can take advantage of previous scheduling activity, and scheduling a plan can become a complex interaction between the planner and scheduler.
The environment of logistics problem also contains uncertainties, and changes dynamically. Airports can be covered in fog. Gas tanks can only be half full. These uncertainties can represent either unforseen problems, or unforseen opportunities. While this paper doesn't describe the recognition of the problems/opportunities, the fact they exist means that when they are recognized the plans/schedules must be able to change. These changes must both be quick, and be biased towards minimizing the disruption of the current plans/schedule. At the least the cost of changing a plan/schedule must be represented. Turning a ship around to pick up a package doesn't occur for free.