Which is More Robust to Gyroscope Bias? A Comparative Analysis of Rule-Based and Machine Learning-Based Attitude Estimation

Attitude estimation methods based on inertial measurement units (IMUs) have been extensively developed using both rule-based filters and machine learning (ML) techniques. Among the various signal errors in an IMU, which includes an accelerometer and a gyroscope, gyroscope bias is a critical factor contributing to estimation errors. Although many studies have focused on developing bias compensation methods, there is a lack of research analyzing the robustness of estimation methods under varying magnitudes of gyroscope bias. This study aims to comparatively analyze the robustness of rule-based and ML-based attitude estimation methods to gyroscope bias. The analysis revealed that for filter-based estimation, the presence of a bias compensation mechanism significantly reduced performance degradation. In contrast, for ML-based estimation, the inclusion of bias augmentation had little impact on performance. Furthermore, when a bias compensation mechanism was applied to the rule-based filter, its performance exceeded those of the ML models. These findings provide insights into the robustness of attitude estimation methods under gyroscope bias.