Compensated heading angles and aerial online mapping system for UGV path generation in magnetically disturbed environments

Abstract

Unmanned Ground Vehicles (UGVs) have performed diverse missions such as delivery, transportation, exploration, and search and rescue. However, UGVs suffer from various disturbances in the real-world. In this dissertation, I focused on three challenges for UGVs: compensation of heading angle distorted in magnetically disturbed environments, compensation of UGV position and velocity in GPS degraded areas, and extension of limited sight of UGV in unstructured environments. I introduce research results for the challenges in three parts. In part 1, I propose an Attitude and Heading Reference System (AHRS) detecting and rejecting magnetic disturbances. The heading angle of vehicles is essential to determine the direction of course for missions. The heading angle measured by a magnetometer is useful to correct the heading angle of the UGV; however, the magnetic field is often distorted by ferromagnetic objects and magnetic sources, causing incorrect estimation of the heading angle. This thesis describes an algorithm to detect and reject magnetic disturbances. The algorithm, based on an Extended Kalman filter, is implemented in a low-cost embedded system. The system is demonstrated in various outdoor environments with magnetic disturbances such as a rooftop of a building, a parking lot, and a sidewalk. The results show that the accuracy of the estimated heading angle is significantly improved as the peak-to-peak error decreases by 32.9%. In part 2, I propose a configuration for the Inertial Navigation System (INS) to improve the position estimation of the UGV in GPS degraded areas by using dynamic constraints. The position measured by GPS is useful to estimate the position of a vehicle, critical information to decide the actions of the vehicle. However, errors of position measured by GPS should be compensated for when the GPS signals are degraded by canopies such as trees and buildings. The proposed system corrects the GPS position error by dynamic constraints, assuming that the UGV is usually not able to move vertically and horizontally perpendicular to the ground and heading angle. The system, implemented by small low-cost embedded sensors on the UGV, is demonstrated on a rooftop of a building with degraded GPS. The results show that the proposed configuration can estimate the precise position even when positions from GPS are distorted and the odometry position measured by the wheel encoders continuously drifts. In part 3, I propose an aerial online mapping system for UGV path generation in unstructured environments. Optimal paths for UGVs should be generated so that vehicles can efficiently perform their missions. Various sensors such as LiDAR, radar, sonar, and camera are used to detect obstacles around the UGV. However, UGV often cannot generate an optimal path due to undetectable obstacles when the sight of the vehicle perception sensors is limited by barriers. The proposed aerial online mapping system provides UGV with a wide view by using aerial images captured by a UAV. Obstacles in the aerial images are detected by a conventional Convolutional Neural Network (CNN) based object detector. The detected obstacles are projected onto the ground in the navigation frame as global obstacles. The system generates a 2D cost map for UGV path generation by merging global obstacles and local obstacles detected by 2D LiDAR. The proposed system is implemented by a quadcopter and 4-wheel skid steering mobile vehicle with a relatively small, low-cost embedded system operating in real-time. I demonstrate the system in a park with undetectable obstacles before the UGV approaches. The results show that the aerial online mapping system can generate an efficient path for the UGV in the real world.