Mobile Robot State Estimation Based on Aided Inertial Navigation

Abstract

Robust and accurate state estimation is a fundamental building block that enables mobile robots to achieve autonomy. To achieve this goal, modern mobile robots are typically equipped with a variety of sensors, such as cameras, light detection and ranging sensors (LiDARs), Global Navigation Satellite System (GNSS) receivers, and inertial measurement units (IMUs), because these sensors provide complementary information about the environment and the robot's egomotion. Among all these sensors, IMUs play a central role as they can work independently without being affected by the environments. Additionally, IMUs can provide high-frequency 6 degree-of-freedom (DoF) motion measurements and are low-cost. However, IMUs are prone to inherent noise and bias instability, resulting in significant error drift over time. <br/>

To mitigate IMU drift, various aiding sources have been integrated with inertial navigation systems (INS). However, exteroceptive aiding sensors are not always reliable: GNSS deteriorates in urban canyons and forests; odometers are sensitive to wheel slip and terrain variations; and cameras and LiDARs are affected by lighting and weather conditions. In addition to these external sensors, the motion profiles of different mobile robots can also be exploited to aid IMUs in a self-contained manner and reduce error drift when other sensors fail. <br/>

Therefore, it is crucial to fuse all available information from different sources to achieve robust and accurate state estimation. On one hand, advanced sensor fusion algorithms must be developed to fully take the complementary advantages of IMU and other sensors. On the other hand, enhancing IMU-based proprioceptive state estimation is essential to ensure continuous and reliable pose estimation when other sensors fail due to environmental disturbances. <br/>

The main contribution of this thesis is the development of novel approaches that exploits different aids to improve the accuracy, robustness, and efficiency of low-cost IMU-based state estimation. These contributions include not only improvements to IMU-based proprioceptive odometry but also the fusion of IMUs with other exteroceptive sensors. Emphasis is placed on the use of low-cost inertial sensors, with the goal of reducing overall hardware costs and thereby enabling large-scale deployment. <br/>

All our proposed approaches presented in this thesis have been published in peer-reviewed conference papers and journal articles. Our proposed LiDAR-inertial odometry system (LIO-EKF) received the second place award in the LiDAR-Inertial Track of the 2023 International Conference on Computer Vision SLAM Challenge. Additionally, we have made implementations of our methods presented in this thesis open-source to benefit the community.