Vision-Aided Autonomous Precision Weapon Terminal Guidance Using a Tightly-Coupled INS and Predictive Rendering Techniques

摘要

This thesis documents the development of the Vision-Aided Navigation using Statistical Predictive Rendering (VANSPR) algorithm which seeks to enhance the endgame navigation solution possible by inertial measurements alone. The eventual goal is a precision weapon that does not rely on GPS, functions autonomously, thrives in complex 3-D environments, and is impervious to jamming. The predictive rendering is performed by viewpoint manipulation of computer-generated of target objects. A navigation solution is determined by an Unscented Kalman Filter (UKF) which corrects positional errors by comparing camera images with a collection of statistically significant virtual images. Results indicate that the test algorithm is a viable method of aiding an inertial-only navigation system to achieve the precision necessary for most tactical strikes. On 14 flight test runs, the average positional error was 166 feet at endgame, compared with an inertial-only error of 411 feet.