Positioning and navigation of wheeled mobile robots in challenging GPS environments

We are witnessing an unprecedented surge of developments in mobile robot navigation after the US government removed selective availability (SA) of the global positioning system (GPS). Although GPS is a good choice for mobile robot navigation it has some drawbacks that make it ineffective for positioning in proximity to buildings, trees and other obstructions.;Localization data from three trajectories is collected for a reference solution, low-cost IMU and wheel encoders. This data is post-processed in an off-line environment to assess the effectiveness of the proposed technique. Three accelerometers and one gyroscope from a low-cost inertial measurement unit (IMU) are processed to obtain positions, velocities and orientations in reduced inertial sensor system (RISS) mechanization. After de-nosing and removal of IMU biases the outputs from RISS mechanization are fed to a Kalman Filter (KF). The KF uses stochastic modeling of IMU errors and predictive modeling of errors from RISS mechanization. The predicted errors are blended with a measurement update in the KF to increase the accuracy of the 3-D estimates for the mobile robot's position, velocity and orientation. The errors between the reference solution and the proposed solution are measured using root means squared (RMS) calculation to determine the proposed solution's effectiveness.;The primary contribution of this thesis is the development of a predictive error model used in KF for estimating the errors in positions, velocities and attitude from RISS mechanization. This thesis will show that this error model when combined with 3-D measurement updates using pitch, azimuth and forward velocity from encoders in KF is a good technique for greatly reducing localization errors.;The secondary contribution from this thesis is based on the equations for pitch, roll and azimuth from prior works [23], [22] and [27] in addition to work done by Jacques Georgy for provide velocities and positions from RISS mechanization. The contribution therefore in this thesis is to take the outputs from RISS mechanization and process them in a Kalman Filter to estimate and correct errors in positions, velocities and azimuth.;In certain situations GPS becomes unreliable or unavailable due to obstructions such as buildings and trees. In absence of GPS, a solution for position estimates with a minimum dollar cost is preferred for small wheeled robots. Unfortunately the low-cost solution contains errors that cause rapid deterioration in position estimates. There must be a way to increase the accuracy of the low-cost solution without adding expensive sensors.;The tertiary contribution from this thesis is a mobile robot. The robot is constructed by the author from nuts, bolts and other raw materials in order to provide researchers with a method for collecting localization data to verify the effectiveness of the proposed technique. The robot is tele-operated over flat and non-flat terrain along the campus roadway, sidewalk and one parking lot of the Royal Military College of Canada.;The length of outages in the experiments varies from 45 to 314 seconds with an average outage length of 134.8 seconds. This outage length is typical for a small mobile robot moving at a speed of 0.5 to 1.0 km/h on a road or sidewalk close to large buildings and clusters of trees. The average RMS error for 3-D positions from KF without measurement updates is 77.6 meters while the average RMS error for KF with 3-D measurement updates is 7.2 meters.;Keywords. Mobile Robot Localization, Kalman Filtering.