Paul DeBitetto, Brent Appleby, Matthew Bottkol, Dale Landis, Spencer Ahrens
Date of Award
Master of Science
Department or Academic Unit
College of Engineering. Department of Electrical and Computer Engineering.
Electrical and computer engineering, Robotics, Visual aid
Vehicles (Remotely piloted)--Design and construction
As UAVs and other unmanned vehicles continue to become smaller, so too does their available processing and sensing capability. For current UAVs, the size, weight, power and cost are all driving factors behind the selection of the onboard avionics, and thus vehicle capability. Nearly all vehicles used in practice today rely on GPS for their navigation abilities. While this is not a compromise in performance for many applications, it does not work well in environments that would benefit most from the use of small unmanned vehicles: indoors, under water, and in dense urban settings. Nearly all small vehicles, by virtue of their intended purpose, already include cameras. In addition, most include inexpensive IMUs, especially those with fast dynamics such as ying vehicles. Even the highest quality IMUs are insufficient for navigation without an external measurement to aid in drift reduction. Likewise, a monocular camera is unable to provide an accurate scale of the environment without some a priori knowledge of the scene. A recent focus in available literature has been on the fusing of these two complementary sensing modalities for use in GPS-challenged navigation. Many approaches are based on the well known Simultaneous Localization and Mapping algorithm. These approaches can provide accurate navigation solutions, but at a high computational cost. This thesis focuses on a vision-aided inertial navigation strategy suitable for small embedded processors found in many of today's unmanned vehicles. A suboptimal covariance analysis and error budget is presented which give a sound theoretical backing to several design decisions related to algorithm efficiency. Analysis for several parameters is presented, including reduced state space, measurement error, and initial uncertainty of the scene. Consideration is given to the numerical dynamic range needed for the Kalman Filter implementation, and matrix factorization methods are employed to ensure numerical stability in the presence of low precision integer math found on many low-cost embedded processors.
Gregory L. Andrews
Andrews, Gregory L., "Implementation considerations for vision-aided inertial navigation" (2008). Electrical and Computer Engineering Master's Theses. Paper 7. http://hdl.handle.net/2047/d10017173
Click button above to open, or right-click to save.