Investigation of Magnetometer-Inertial SLAM for Autonomous Railway Robot Navigation

Abstract

In GPS-denied railway environments, such as tunnels or underground stations, estimating the speed of a moving platform remains a challenge. This thesis presents a sensor-based approach using magnetometers and inertial measurement units (IMUs) to estimate velocity without relying on external infrastructure. We first simulate realistic magnetometer and IMU data based on motion profiles, adding noise such as Gaussian jitter, temperature drift, spatial jitter, and random spikes. The synthetic magnetic field is created using a combination of sinusoidal functions to reflect real-world magnetic variations. Velocity is then estimated by comparing signals from front and rear magnetometers. By measuring the time delay between them, speed can be calculated using the known distance between sensors. The method includes filtering and window-based lag detection for stability under dynamic conditions. The estimated speed is fused with IMU acceleration using an Extended Kalman Filter (EKF). Results show that IMU-only fusion performs well, but direct use of noisy magnetometer speed may degrade accuracy. To improve this, we propose a complementary filter as a pre-fusion step before EKF. This work demonstrates a lightweight and effective method for speed estimation in low-GPS or GPS-free railway scenarios. Future work will include improving signal quality, implementing adaptive fusion, and hard-in-loop testing in real robots.