|M.Sc Student||Rothman Yaakov|
|Subject||Mitigating Navigation Drift using Optical Sensors and|
|Department||Department of Civil and Environmental Engineering||Supervisors||Professor Sagi Filin|
|Dr. Itzik Klein|
|Full Thesis text|
Navigation has been in common use for many years, ranging from door to door vehicle navigation, automated processes, and mapping technologies. The most common form of navigation nowadays is via point positioning using global navigation satellite system (GNSS) receivers. However, this method requires a clear line of sight to at least four satellites, which can be negated in urban settings, underground or indoors.
Inertial navigation systems (INS) have an advantage over GNSS based ones, as they rely on measurements derived from the platform itself, and not externally. However, inherent biases in the inertial sensors cause the navigation solution to drift overtime. To overcome this, aiding schemes have been proposed where INS and GNSS are combined. This has the same constraints as a standalone GNSS solution and therefore inapplicable in the settings described.
In order to reduce inertial navigation drift, vehicle constraints have been gaining increased popularity as a means to mitigate drift of inertial navigation systems in GNSS deprived settings. Another method, which is popular for mapping technologies is simultaneous localization and mapping, which also provides a three dimensional model of the surroundings. While useful, such constraints cannot compensate for the drift of all state variables in the navigation solution. To study which variables are affected, empirical analyses under typical scenarios are commonly performed, however, no insight is provided into the inner effects amongst error states.
This thesis develops analytical observability terms which evaluate the contribution of vehicle constraint measurements to the solution and determines which error states or linear combination of them are unobservable. Observability analysis has been mostly directed thus far at GNSS/INS aiding and is generally performed in a binary manner.
In this context this thesis fills this gap by deriving analytical terms for the unobservable subspace of the body velocity constraint, which is already utilized in urban driving scenarios. Additionally, an approach to mitigate inertial drift combining simultaneous localization and mapping and the body velocity constraint is proposed. For this combination, analytical terms of the unobservable subspace are derived as well. In this approach we show that it provides full error state observability for most urban driving schemes.
The analytical terms derived in this research are verified via a numerical analysis of urban driving scenarios using a degree of observability approach. This approach is normalized as compared to a regular covariance analysis, and therefore the results are not dependent on initial accuracies of the error states.