Beyond the Plan-Length Criterion

Resources such as food, hit points or magical power are standard features of today's computer games. Computer-guided characters often have goals that are related to these resources and must take this into account in planning their behavior. In the eld of artiicial intelligence , resource-based action planning systems have recently begun to mushroom. However, these systems still tend to neglect resources. Conventional plan length is the primary optimization goal, resources being of only secondary importance. This paper illustrates the problem and gives an overview of the Excalibur agent's planning system, which overcomes these deeciencies by applying an extended constraint programming framework to planning. It allows the search space to be explored without the need to focus on plan length, makes it possible to optimize other criteria like resource-related properties and promotes the inclusion of domain-dependent knowledge to guide and accelerate the search for a plan.