Empirical Analysis of Sensor Fusion Algorithms for Improved Localization Accuracy in Autonomous Mobile Robots Operating in Unstructured Nigerian Industrial Environments

Autonomous mobile robot localization in unstructured and GPS-denied industrial environments such as Nigerian manufacturing facilities poses significant technical challenges due to featureless layouts, sensor noise, and dynamic obstacle populations. This study presents an empirical analysis of three sensor fusion algorithms, extended Kalman filter, unscented Kalman filter, and particle filter, for improving localization accuracy in a wheeled mobile robot operating within an industrial warehouse environment in Aba, Abia State. A differential-drive experimental robot platform is equipped with a 2D LiDAR scanner, wheel odometry encoders, and an IMU, providing complementary sensing modalities for fusion. Each algorithm is implemented on a Robot Operating System framework running on an onboard NVIDIA Jetson Nano single-board computer. Localization accuracy is evaluated by comparing estimated robot poses against ground-truth positions obtained via an OptiTrack motion capture reference system installed in the test environment. Experiments are conducted across three environmental conditions representing low, medium, and high obstacle density. Results indicate that the particle filter achieves the lowest mean position error of 3.8 centimetres in high-obstacle-density conditions, outperforming the EKF at 6.1 centimetres and UKF at 5.2 centimetres, though with higher computational cost. The trade-off between localization accuracy and computational resource consumption is quantified for each algorithm, providing selection guidance for deployments with constrained onboard processing capacity. The study fills an identified gap in sensor fusion benchmarking conducted under Nigerian industrial spatial conditions. Keywords: sensor fusion, mobile robot localization, Kalman filter, particle filter, autonomous robot