The control of autonomous underwater vehicles through a hierarchical structure of value priorities

A value-driven control methodology is being developed to accomplish the higher-level decision and planning functions for the group of cooperating Autonomous Underwater Vehicles (AUVs) that are to be demonstrated in Lake Winnipesaukee. The control structure, which is being developed jointly by Decision-Science Applications and the National Bureau of Standards, follows the same hierarchical structure that has been used by the Bureau of Standards for automated manufacturing and robotic control, but it incorporates the DSA value-driven control methodology to provide a flexible decision paradigm for the upper levels of the hierarchy. The planned scenarios are designed to demonstrate the "intelligence" and flexibility of the control system in balancing stealth versus other mission objectives in the choice of tactics and routing alternatives for a potentially hostile environment. The talk will discuss the specific structure of the decision processes and will show examples of system performance as derived from test examples. Previous experience has shown that a hierarchical structure of value priorities can provide a flexible way of controlling multi-layered automated and semi-automated systems. The most comprehensive existing application is in the context of a detailed multi-flight air combat simulation, known as TAC BRAWLER, which provides a fully-automated representation of all of the C
³
I and decision processes for both the Red and Blue flight leaders and pilots. In detailed comparisons with field exercises and man-in-the-loop simulations, the TAC BRAWLER model has been found to provide a realistic simulation of human combat behavior-including even key "human factors" such as surprise, confusion, and mental overload. In a value-driven control hierarchy, the decision alternatives at each level are scored, or evaluated, in terms of a pre-defined set of valuative "considerations" (or sub-objectives). The weight or importance for each consideration in the decision process is governed by a scalar "value" that is specified dynamically at the next higher level in the control hierarchy. Thus, the "policy intent" at each level in the hierarchy is communicated through a simple "vector of values" to the next lower level in the control hierarchy. This simple method of communicating the policy intent allows such systems to exhibit a remarkably flexible responsiveness to user priorities that can be dynamically specified at the top by the user.