Attitude Estimation with an Invariant Extended Kalman Filter Using Learning-Based Covariance Adaptation

Date

2025-9-1

Author

Arslan, Mehmet Emir

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

228
views

0
downloads

Cite This

This thesis proposes a data-driven method to improve attitude estimation performance by adaptively scaling the measurement noise covariance matrix within an Invariant Extended Kalman Filter (IEKF) framework. To achieve this, a convolutional neural network (CNN) adjusts the measurement covariance based on recent sensor data, allowing the filter to adapt its reliance in measurements under different motion and environmental conditions. The CNN processes sequences of inertial and magnetic sensor readings, including gyroscope, accelerometer, and magnetometer measurements. This adaptive mechanism allows the IEKF to better handle challenging scenarios such as translations, fast rotations and magnetic disturbance cases. The effectiveness of the method is evaluated on publicly available BROAD dataset. Experimental results demonstrate improved estimation accuracy not only over the fixed-covariance IEKF, but also compared to IEKF with conventional adaptive methods.

Subject Keywords

Attitude Estimation, Invariant Extended Kalman Filter, Adaptive Covariance Estimation, Convolutional Neural Networks

URI

https://hdl.handle.net/11511/116152

Collections

Graduate School of Natural and Applied Sciences, Thesis