You're invited to hear a

Ph.D. Thesis Defense

by

Kyuman Lee

(Advisor: Professor Eric N. Johnson)

"Adaptive Filtering for Vision-Aided Inertial Navigation"
 

11:00 AM, Monday, November 5
Montgomery Knight Building Room 317

Abstract:
With the advent of unmanned aerial vehicles (UAVs), a major area of interest in the research field of UAVs has been vision-aided inertial navigation systems (V-INS). Many missions of UAVs— reconnaissance, damage assessment, exploration, and other guidance, navigation, and control (GNC) tasks—often demand V-INS in more operational environments such as indoors, hostilities, and disasters. In V-INS, inertial measurement unit (IMU) dead reckoning generates the dynamic models of UAVs, and vision sensors extract information about the surrounding environment and determine features or points of interest. With these sensors, the most widely used algorithm for estimating vehicle and feature states of V-INS is an extended Kalman filter (EKF). The design of the standard EKF does not inherently allow for time offsets between the timestamps of the IMU and vision data, and the necessary assumption of the EKF is Gaussian and white noise. In fact, sensor-related delays and measurement outliers that arise in various realistic conditions are unknown parameters. A lack of compensation of unknown parameters leads to a serious impact on the accuracy of the navigation systems. To compensate for uncertainties of the parameters, we require modified versions of the estimator or the incorporation of other techniques into the filter.

The main purpose of this thesis is to develop reliable and robust V-INS for UAVs, in particular, those for situations pertaining to such unknown parameters. First, to fuse measurements with unknown time delays, this study incorporates parameter estimation into feature initialization and state estimation. Utilizing estimated delays and cross covariance, online temporal calibration, called “latency-adaptive filtering,” corrects residual, Jacobian, and covariance. In addition, feature correspondence in image processing front end rejects vision outliers, and then a chi-squared statistic test in filtering back end detects the remaining outliers of the vision data. For frequent outliers, variational approximation for Bayesian inference derives how to compute the optimal noise precision matrices of the measurement outliers. These overall processes of outlier removal and adaptation refer to as “noise-adaptive filtering.” Even though almost all of V-INS remove outliers by their own methods, unfortunately, few researchers have treated outlier adaptation in V-INS in great detail. Results from flight dataset tests validate the improved accuracy and reliability of V-INS employing these adaptive filtering frameworks.

Committee Members:
Professor Eric N. Johnson, Dept. of Aerospace Eng., The Pennsylvania State University (Advisor)
Professor Eric Feron, School of Aerospace Engineering (Co-Advisor)
Professor E. Glenn Lightsey, School of Aerospace Engineering
Professor Marcus J. Holzinger, Aerospace Eng. Sciences, University of Colorado Boulder
Professor Byron Boots, School of Interactive Computing