Development of a Data Collection System for Tightly Integrated GNSS, IMU, Radar, and LiDAR Navigation

There is a growing interest in autonomous driving systems that can safely rely on multiple sensors including GNSS, IMU, Radar and LiDAR to navigate with high accuracy, integrity, continuity, and availability in complex urban environments. Many existing data sets, collected with multi-sensor platforms, focus on validating different variations of visual localization algorithms like SLAM, place recognition, object detection and visual odometry that help navigate in sky-obstructed and GNSS-denied environments. However, GNSS still plays a vital role in providing the most assured navigation solution. In this thesis, we develop a robust system intended for collecting data sets that will support the design of tightly integrated navigation algorithms and the analysis of integrity risk using GNSS coupled with IMU, Radar, and LiDAR in challenging automotive environments. GNSS pseudorange, doppler, and carrier phase and IMU acceleration and angular velocities are measurements that the system is specifically designed to collect for sensor-fusion algorithm refinement. In addition, time synchronization between sensors is crucial in data sets validating tightly integrated navigation, especially in applications with high dynamics. However, there is no widely accepted accurate and stable method for synchronizing clocks between different sensor types. We implement a common-clock synchronization and a hardware-trigger clock synchronization between multiple sensors. We then collect a preliminary data set to compare the accuracy and stability of sensor time-tagging using a GNSS-receiver-generated hardware trigger versus using a local-clock ROS-based time stamping. We evaluate the impact of these synchronization methods on mapping accuracy performance.