Extracting Plans From Plans

Propositional plans of all forms often display a certain level of concurrency which can be exploited by scheduling the plan. This reveals the earliest times at which each action can be applied whilst still achieving the goal and shortening the plan duration. However, the output of this scheduling process is simply a set of timestamped actions, losing implicit information present in the original plan such as the causal links between actions and states. In this paper we present PIMP (Plans Inside Multi-threaded Plans), a domain-independent algorithm which can schedule a plan whilst retaining the knowledge inherent in a traditional plan. We exploit this using the concept of threads to detect individual, concurrent and interleaved plans and discuss the benefits of these thread-scheduled plans and their possible