An Expert Distributed Robotics System with Comprehension and Learning Abilities in the Aircraft Flight Domain

Abstract : The authors are focusing on in-flight problem diagnosis. Suppose, for example, a pilot simultaneously experiences over-heating in one engine and aileron reverse. He might attribute the problem to the hydraulic system, but unless he possessed detailed technical knowledge of the particular aircraft, he might not be able to decide which sub-assembly component became disfunctional. However, exactly how and where the problem occurred may have implications for how to deal with it. Indeed, a naively plausible, but wrong, assessment of the problem may lead the pilot to exacerbate rather than improve his situation. On-board intelligent computer system to aid in diagnosis and to suggest corrective measures would be of great help. There are a number of essential attributes of such a system. Such a system raises a number of theoretically important issues of interest to the investigators. Very broadly, the issues are: (1) developing computer representations for physical mechanisms, (2) intelligent modeling of those mechanisms, (3) high level natural language communication between humans and computers, and (4) learning from experience and instruction. The remainder of this report is divided into three sections. Each describes the progress reported by one investigator. Several sections contribute to more than one of the four facets of the in-flight diagnosis problem (representation, mechanism modeling, natural language, and learning).