Partial Order Planning with Goals of Prevention

In real world domains, it is often necessary to avoid some undesirable situations, or intermediate states, from occurring during a plan. For example, when planning the operation of a chemical plant, it is necessary to avoid creating unsafe mixtures of chemicals. This paper presents an algorithm for explicitly preventing unsafe situations from occuring during a plan. The algorithm is presented as an alternative to protecting the safety of a plan by rewriting the operators in the planning domain. The latter approach is described in Weld and Etzioni (1994). The algorithm described here has two main procedures. The rst procedure monitors the safety of the plan. It is called at the end of every planning cycle. The second procedure is called when the safety of the plan is under threat. It applies plan modiication rules to correct plan safety. The aproach described in this paper has two advantages. Firstly, each modiication that is made to protect the safety of the plan is made because the plan would otherwise be unsafe. This is important if the planner is expected to explain the decisions it makes during planning. Secondly the planner idea of safety is deened by the algorithm to monitor the safety of a plan. Complex ideas of safety can be described by modifying this algorithm. For example, the algorithm can be adapted to acknowledge certain situations as unsafe only during certain regions of time.