|dc.description.abstract||This thesis presents an inertial navigation system (INS) that leverages the global positioning system (GPS) and a sparse road network database to cooperatively localize ground vehicles within close proximity to one another. The algorithm that constitutes the core contribution is named MACIN, an acronym for map aided cooperative inertial navigation.
Increasing demand for driver assistance features in consumer ground vehicles has spurred demand for ubiquitous high-accuracy absolute positioning. Accuracy at the lane level (under 1 meter) is required to execute complex operations such as maneuver planning. At the same time, standard automotive sensors such as cameras, inertial measurement units (IMUs), and GPS receivers do not provide this accuracy. Furthermore, common navigation techniques for fusing these sensor measurements, such as loose GPS/INS coupling in an extended Kalman filter (EKF), produce position performance that is consistently on the order of several meters in benign conditions.
MACIN comprises several improvements upon the loosely coupled GPS/INS EKF approach to acheive sub-meter accuracy and accurate lane determination. It uses sparse lane geometry information and lane sensing capability to apply position constraints along the earth tangent plane. The states of neighboring vehicles are estimated concurrently, and differential GPS is used to relate their states to one another. Lastly, Rao-Blackwellized particle filtering (RBPF) is used to estimate position with particles, while all other variables within the state are estimated with standard linearized filtering.
The success of these improvements is measured by reduction of positional error along the earth tangent plane. MACIN’s performance is compared to that of a loosely coupled GPS/INS EKF in both highway and suburban conditions. This thesis shows that the proposed novel filter consistently reduces error from 1-3 meters to the submeter level.||en_US