Improving Magnetic Map-Based Navigation using Vehicle Motion Information

Abstract

This thesis utilizes the earth's main magnetic field as the signal for a map definition, implements a particle filter to synthesize a solution from likelihood information, and introduces vehicle motion information by means of velocity and heading measurements to augment the traditional filter structure. Three different process models are investigated: Gauss-Markov, a known capable technique in which particles evolve stochastically; wheel speed linear motion with magnetometer heading updates, in which the particles attempt to imitate the movement of the vehicle using a heading derived from magnetic north (with declination correction); and wheel speed linear motion with gyroscope heading updates, in which angular velocity measurements are integrated to update the heading at every time step instead. Measurement updates were performed with respect to the map and the local magnetic signal. Two nominal routes were evaluated using the Gauss-Markov approach as a baseline and the root-mean-square error in the position estimate compared to that of the motion-informed models. Results show that vehicle odometry can decrease the error in the position solution by between 22% and 77% on average and decrease the typical maximum error along a route by about 54%. This focus was expanded to explore non-nominal driving conditions including a map driven in reverse, a map featuring a short detour, and a map featuring a large detour. In these cases, every filter implementation struggled to track the vehicle once it exited the map, but recovery was possible in some instances when the vehicle returned to a known or strongly identifiable region.