Priority-Based Meta-control within Hierarchical Task Network Planning

摘要

We are developing real-time planning and control systems that allow a single human operator to control a team of unmanned aerial vehicles (UAVs). If the operator requests more tasks than can be immediately addressed by the available UAVs, our planning system must choose which goals to try to achieve, and which to postpone for later effort. To make this decision-making easily understandable and controllable, we allow the user to assign strict priorities to goals, ensuring that if a goal is assigned the highest priority, the system will use every resource available to try to build a successful plan to achieve that goal. In this paper we show how unique features of the SHOP2 hierarchical task network planner permit an elegant implementation of this priority queue behavior. Although this paper is primarily about the technique itself, rather than SHOP2's performance, we assess the scalability of this priority queue approach and discuss potential directions for improvement, as well as more general forms of meta-control within SHOP2 domains.