ENHANCED MODELING AND NEURAL NETWORKS FOR AUTONOMOUS AIR VEHICLE OPERATIONS

THE FEASIBILITY OF AN ALTERNATIVE APPROACH TO AUTONOMOUS OPERATION OF AN UNMANNED AIR VEHICLE (UAV) IN A RSTA ENVIRONMENT WILL BE STUDIED. THE APPROACH EMPLOYS AN ENHANCED MODEL OF THE SOFTWARE "FLIGHT CREW" COUPLED WITH ADAPTIVE LEARNING ALGORITHMS. A SYSTEM IS PROPOSED THAT WILL DECENTRALIZE CRITICAL UAV C2 FUNCTIONS AND OPTIMIZE COMMUNICATIONS, SUCH THAT GROUND STATION WORKLOAD/MANPOWER IS MINIMIZED, THE VEHICLE IS ALLOWED TO OPERATE IN ENEMY'S DEEP WITHOUT CONSTANT DATA-LINK, AND THE VEHICLE CAN REACT TO DANGEROUS OR OPPORTUNISTIC SITUATIONS WITHOUT THE DOWNLINK, GROUND STATION REPLAN, UPLINK DELAY. THE STUDY WILL INCLUDE AN ANALYSIS OF THE PARAMETER SPACE REQUIRED FOR AN EFFECTIVE REAL-TIME NEURAL NETWORKBASED SYSTEM AND A MORE NATURAL GROUND STATION/UAV DIVISION OF LABOR (BASED ON MANNED FLIGHT CREW ARCHITECTURE). THIS WILL ENHANCE SURVIVABILITY AND THE ACCOMPLISHMENT OF MISSION OBJECTIVES. THE ADA-BASED SYSTEM WILL BE COUPLED TO AN EXISTING GROUND-BASED MISSION PLANNER (COMMON LISP ON IBM AT) THAT PROVIDES FLIGHT PROFILES FOR RSTA MISSIONS IN THREAT ENVIRONMENTS.