A Novel Approach for IMU Denoising using Machine Learning

Inertial Measurement Unit (IMU) sensors are used predominantly in navigation systems. Micro-Electro-Mechanical Systems (MEMS) sensors have been introduced as a cost-effective lightweight IMU. However, MEMS IMU has larger stochastic errors that accumulate over time, causing navigation drifts. This issue is dealt with by fusing IMUs with Global Navigation Satellite System (GNSS) to obtain reliable navigation. This fusion setup fails to provide continuous, reliable navigation during GNSS outage scenarios due to IMU errors. So in this paper, we propose a novel approach to reduce the navigation drifts by removing IMU errors using Light Gradient Boosting Machine (LightGBM) Machine Learning algorithm. Unlike many other works that use high-end expensive IMU to train the model to denoise low-cost MEMS IMU, this paper uses Inverse Kinematics to obtain clean IMU training data from the Position, Velocity and Attitude (PVA) values estimated using Extended Kalman Filter (EKF) when the GNSS is available and reliable. The proposed method is tested in both simulation and real data sets under different GNSS outage durations. Results showed significant improvement in position, velocity and orientation estimation.