Scalable High Performance Computing Using Autonomous Agents

The study of "deterministically chaotic" systems is embryonic and its application to real-time simulation and control is just beginning. A process is determininistically chaotic, as opposed to random if its current value depends on its previous values, but its current value is not predictable. System which have significant non-linear aspects exhibit some degree of chaotic behaviour and in large, complex manufacturing systems, the system may be significantly non-linear. Current tools for planning, scheduling, and simulation for large manufaturing systems generally attempt to use linear mathematics to approximate the system under consideration, and as a result are complex, and brittle. At least one reason for this poor performance is that with chaotic systems, attempts to use closed form mathematics to optimize answers may be hopeless. Under the work herein proposed we will develop both the theory and practice of applying autonomous agents to planning, scheduling, and simulation within a complex manufacturing facility. We will demonstrate this system at Watervliet Arsenal, where our prototype system will provide superior results orders of magnitude faster, with less complex software, and with greater flexibility and reliability than commercially available tools. Computer systems based on autonomous agents are inherently parallel and can execute on massively parallel hardware with no specific parallization. Software and hardware based on the paradigm of autonomous agents are also inherently scalable. Anticipated Benefits: Our work provide the foundation for an entirely new class of planning, scheduling, and simulation systems for manufacturing applications. Beyond manufacturing, our approach to planning and scheduling could be applied in mnay other application areas. As an example, we have demonstrated the use of autonomous agents for battle management, and our initial results are very encouraging.