Planning Tactics within Scheduling Problems

In this paper we consider the possibilities and potential advantages for exploiting automated planning techniques in the service of solving scheduling problems. The core competency of scheduling technologies is allocation of resources to pre-specified networks of competing activities (typically belonging to multiple processes) to maximize aspects of global system behavior. However, in practical domains it is rarely the case that the problem can be treated strictly as an allocation problem (i.e., strictly as a problem of enforcing disjunctive resource constraints in conjunction with specified temporal constraints). Rather some level of dynamic action selection is invariably required, typically to transition resources from one usage to the next. To retain scalability, schedulers tend to make use of locally circumscribed assumptions about the dynamics of resource usage that fit the problem at hand, which allows for efficient generation of resource-support plans without explicit reasoning about goals. But, these approaches can be overly restrictive in many cases, and they also tend to be difficult to extend and reuse.