Integrating Partial-Order Planning

Reasoning about plans and actions is fundamental to intelligent control of real-world systems such as AUVs. These plans can come from a library of existing plans or from creating a new plan if no appropriate plan exists for the situation. The Orca project [Turner, 1994] is an ongoing effort to create an intelligent system for mission-level control of autonomous underwater vehicles. Orca’s primary means of achieving goals is to retrieve existing plans from memory and to apply them by doing their steps. This paper describes our effort to integrate partial-order planning into Orca’s schema-based planning mechanism. With this work, Orca first looks for an appropriate plan in its repertoire of plans to achieve a goal or a set of goals. If such a plan is not available, it considers more general plans. For some goals, even a generalized plan may not be readily available. In such cases, Orca uses partial-order planning to generate a specific plan “from scratch” for the problem-solving situation. This paper describes our implementation of such a planner and its seamless integration into Orca’s schema-based reasoning mechanism. Orca is now capable of creating new plans if necessary, criticizing and patching a plan (if needed) just prior to execution of the plan, and handling conjunctive, disjunctive, and negated goals.