Sensor Fusion on Embedded Platform for Holonomic Indoor Robot Navigation

Abstract

Accurate self-localization is a critical requirement for effective navigation and obstacle avoidance in indoor holonomic robotic systems. This paper presents an experimental evaluation of multiple localization techniques, focusing on the fusion of wheel encoder and Inertial Measurement Unit (IMU) data on a custom-built Mecanum-wheel indoor robot. The study investigates the performance of individual sensor-based localization and compares it with fusion-based approaches employing Extended Kalman Filter (EKF), Unscented Kalman Filter (UKF), and Complementary Filter algorithms, all implemented in real time on an STM32G474RE microcontroller. Key performance indicators (KPIs), including linear and angular RMSE relative to a manually measured ground truth, are analyzed across a square reference trajectory. Results demonstrate that EKF achieves the highest position accuracy (0.09% $x$-error, 0.59% $y$-error), while the Complementary Filter achieves near-zero yaw error (0.02%). These findings provide practical guidelines for selecting sensor fusion algorithms for embedded indoor robot navigation.