A Real-Time Knowledge Processing Executive for Army Rotorcraft Applications

Efforts such as the Army's Rotorcraft Pilot's Associate (RPA) Advanced Technology Transition Demonstration (ATTD) Program have requirements for real-time execution of knowledge-based systems. However, existing computer hardware platforms and software systems (languages, tools, shells, etc.) are not adequate for applications requiring real-time knowledge-based processing. This is not surprising since knowledge-based systems have historically been developed with non-real-time applications as their primary focus (e.g., expert medical consultants, off-line diagnostics, advisors, etc.). Further, these applications generally were not concerned with run-time efficiency, optimal use of computer resources, embedded system requirements (size, weight and power) and other considerations important in addressing real-time processing. The fundamental design concepts necessary for real-time performance were not considered in the design of the languages, shells and tools used for knowledge-based processing. This paper describes research in the development of a real-time knowledge-based system execution environment suitable for use in next-generation Army helicopters. The paper first describes the unique problems which must be solved in real-time knowledge processing applications. We then describe an overall processing architecture for achieving the requisite real-time performance. The execution environment is being specifically designed for embedded applications such as the mission equipment package of the RAH-66 Comanche helicopter. The environment will provide the requisite software execution facilities which heretofore have not been available in an integrated embedded execution environment. Further, this execution environment will provide the necessary control features required to manage real-time problem solving. What is Real-Time AI Processing? Traditionally, computer science has defined a real-time system as a system which must not only provide the correct answer but provide it within strict time constraints. As we shall see, this definition is not adequate for AI processing systems. Al researchers have no generally accepted definition of real-time AI. Many have assumed that faster execution is an acceptable method for attaining real-time performance. To that end, previous approaches to real-time AI have concentrated on speed-up of the process. This speed-up has been accomplished in a number of ways. Solutions tried include: • Separation of development and run-time environments. • Use compiled rather than interpreted code. • Use a faster language, e.g., C rather than LISP. • Use special purpose hardware, e.g., associative processors, LISP machines and parallel architectures.