IRI-9530915 Venkataraman, Subramanian Yale University $75,692 - 12 mos This research addressed the integration of planning and control in intelligent systems from a systems theoretic viewpoint. In doing so, it investigates the relevance of two ideas to integration of planning and control. First, it studies the relationship between a plan element (or action or behavior) and the concept of a reference model in control theory. This, in conjunction with the arsenal of tools available in control theory for the synthesis and analysis of controllers that guarantee reference model following, may then be used to formally specify the control performance of each action. The second aspect of this research deals with the existence of multiple actions in a plan, connected together through some relationships, and having some ordering. To address this, the use of multiple reference models (in a control system) and means to switch between them in a stable fashion is investigated. By studying these problems, an integrated view of certain elements of planning and control may be derived, where reference models are related to actions, switching rules to the initiation and termination of actions, switching sequences to the admissible output of plan executors in deliberative systems, etc. In this manner, the planned work would complement ongoing planning research in AI by providing the control oriented knowledge necessary for robust design. The theoretical results developed in this effort will be applied to quadruped locomotion (as an example problem), and a kernel of gait controllers, and the set of admissible rules for gait switching and tuning that guarantee stable locomotion will be formally derived.