Data fusion for localization using state estimation and machine learning

Abstract

Using multiple sensory information is acknowledged as one of the major topics in the navigation
\nof aerial and ground vehicles. This doctoral thesis considers localization as a state
\nestimation problem, which is solved by data fusion techniques and supported by machine
\nlearning methods. It attempts to address the issue of developing a better localization
\nsystem for a ground vehicle by seeking the best possible pose estimator (i.e., position,
\nvelocity and attitude) and improving its robustness to unexpected sensor measurements.
\nThe vehicle of interest is represented by a skid-steer tracked mobile robot; however, all
\nthe algorithms work with a sensory set, which can be with minor changes deployed on any
\nvehicle, legged, wheeled, tracked, or aerial. First part of this thesis explores the development
\nof different state estimation architectures, which exploit the extended Kalman filter
\nfor full 3D dead reckoning (i.e., incremental or relative pose estimation). The purpose of
\nthis part is to use inertial and odometry dead reckoning to its optimal extent, considering
\nboth the performance and computational complexity. Such combination of proprioceptive
\nsensory modalities used on a ground vehicle is expected to provide the core localization—
\nfoundation for any other higher level localization or navigation systems. Second part of
\nthe dissertation investigates means of improving overall robustness and performance of the
\nmulti-modal state estimation. Different sensory modalities are prone to various types of
\nerrors, especially in an environment that changes dynamically. Therefore, the thesis shows
\nthe importance of identifying and rejecting unexpected or erroneous measurements. The
\nmulti-modal data fusion is based on inertial and odometry measurements aided by information
\nfrom a camera and laser range finder. These two exteroceptive modalities are in
\nparticular prone to real-world disturbances, therefore, they are the subject of anomaly detection
\nprocess. Various state-of-the-art machine learning methods (i.e., logistic regression,
\nSupport Vector Machines, Gaussian Mixture Models and Gaussian Processes) are applied
\nin a Kalman filter framework to monitor the measurements and overcome the commonly
\nused covariance monitoring and chi-squared gating test. Verification of all the techniques
\nin this thesis is supported by extensive experimental datasets collected with a real mobile
\nrobot in both indoor and challenging outdoor environments.