Orca : Intelligent Adaptive Reasoning for Autonomous Underwater Vehicle Control ∗

Real-world problems demand adaptive problem solvers that can tailor their behavior to their task domain, both during a problem-solving session and over time across sessions. Otherwise, incomplete knowledge, uncertainty, the presence of unpredictable agents and processes, and hardware failure and imprecision will conspire to cause unanticipated events, mission failure, and possibly damage to the agent. In this paper, we discuss Orca, a schema-based, context-sensitive adaptive problem solver. Orca is an intelligent mission controller for autonomous underwater vehicles (AUVs). Orca uses procedural schemas, which are like hierarchical plans, to control its actions. It uses contextual schemas, which are similar to generalized cases, to ensure that its behavior is tailored to its problem-solving situation. Two kinds of adaptation are discussed, short-term adaptation to meet the needs of the agent’s current mission, and long-term adaptation, which changes the agent’s knowledge to become better over time at performing all its missions. ∗The author is grateful to the National Science Foundation for grant BCS–9211914, which supported this work in part. The author is also affiliated with the Department of Computer Science at the University of New Hampshire. Correspondence should be addressed to the author at Department of Computer Science, Kingsbury Hall, UNH, Durham, NH 03824. Orca: Intelligent Adaptive Reasoning for Autonomous Underwater Vehicle Control Roy M. Turner Marine Systems Engineering Laboratory Marine Science Center Northeastern University Problem-solving systems designed to operate in the real world need to be adaptive if they are to be of any use. They must be able to handle unanticipated events and change the way they operate to meet the changing demands of their environments and task domains and to mitigate the effects of uncertainty, incomplete knowledge of the environment, and lack of domain knowledge. Adaptation of a problem solver to its problem-solving environment can be broken into two types based on the time over which adaptation takes place (see, e.g., [10]). Short-term adaptation is how the agent changes its behavior to fit its problem-solving environment during the course of a single mission. For long-term missions, there may be some learning involved, but in general, other adaptive mechanisms will be needed. Long-term adaptation involves learning and tailors the agent’s problem-solving knowledge to better fit its overall task domain. One can think of long-term adaptation tuning over time the knowledge an agent uses for short-term adaptation. This paper describes work on the Orca project [10; 13] addressing adaptive reasoning in a real-world domain. Orca is an intelligent, adaptive mission controller for autonomous underwater vehicles (AUVs) that is being built at the University of New Hampshire and Northeastern University. Orca is designed to be a robust controller for AUVs for ocean science missions. Since Orca must work in a real-world domain, we have little choice but to make it adapt, both in the short and long terms, to the exigencies of its problem-solving environment. We are currently focusing on issues related to short-term adaptation, including: run-time flexibility of procedural knowledge, conditioning of behavior via “behavioral parameter” settings, and context-sensitive reasoning via the use of a priori (experiential) contextual knowledge. Ultimately, we will focus on long-term adaptation, including case-based reasoning and other learning mechanisms. In the remainder of this paper, we discuss Orca and its domain, then look at current work on short-term adaptation. We then discuss plans for long-term adaptation, and conclude with a discussion of the project’s current status and future work.