Robust Attitude and Heading Estimation under Dynamic Motion and Magnetic Disturbance.

Area of Science:

• Robotics and Navigation
• Sensor Fusion
• Machine Learning for Sensor Data

Background:

• Accurate attitude and heading estimation is critical for Micro-Electromechanical System (MEMS) Inertial Measurement Units (IMUs) in applications like pedestrian dead reckoning (PDR), human motion tracking, and Micro Aerial Vehicles (MAVs).
• Low-cost MEMS-IMUs suffer from noise, dynamic motion, and magnetic disturbances, compromising Attitude and Heading Reference System (AHRS) accuracy.
• Existing methods struggle with the inherent inaccuracies of MEMS-IMUs in real-world conditions.

Purpose of the Study:

• To develop a novel data-driven IMU calibration model to enhance the accuracy and robustness of attitude and heading estimation.
• To address the challenges posed by sensor noise, external accelerations, and magnetic disturbances in MEMS-IMUs.
• To improve the performance of downstream applications reliant on precise motion tracking.

Main Methods:

• A data-driven IMU calibration model utilizing Temporal Convolutional Networks (TCNs) to denoise sensor data by modeling random errors and disturbances.
• An open-loop, decoupled Extended Complementary Filter (ECF) for robust sensor fusion and accurate attitude estimation.
• Systematic evaluation using three public datasets (TUM VI, EuRoC MAV, OxIOD) with diverse hardware and motion conditions.

Main Results:

• The proposed TCN-based calibration and ECF fusion method demonstrated superior performance compared to advanced baseline methods.
• Achieved over 23.4% improvement in absolute attitude error and 23.9% in absolute yaw error against state-of-the-art techniques.
• Generalization experiments confirmed the model's robustness across different IMU devices and usage patterns.

Conclusions:

• The novel data-driven approach effectively mitigates noise and disturbances in MEMS-IMUs, leading to significantly more accurate attitude and heading estimation.
• The proposed method offers a robust and generalizable solution for enhancing AHRS performance in various challenging environments.
• This work provides a significant advancement for applications requiring precise motion tracking and navigation.