Quaternion-Based Orientation Filter Benchmarking with 9-DOF MARG Sensors: Evaluation via Robot-Guided Motion and MoCAP

Accurate orientation estimation using 9-DOF Magnetic, Angular Rate, and Gravity (MARG) sensors is essential for robotics, unmanned aerial vehicles (UAVs), and motion analysis. This paper presents a comparative evaluation of six quaternion-based orientation filters: Complementary filter, Madgwick filter, 6-DOF and 9-DOF Error-State Kalman Filters (ESKF), the proprietary XKF3i, and the Versatile Quaternion-Based Filter (VQF). Experiments were conducted on a UR3 collaborative robot carrying an Xsens MTi-300 sensor, with reference orientation provided by an OptiTrack motion capture system. Eight robot-guided scenarios, ranging from static drift to dynamic multi-Axis motions, were executed to ensure repeatability and systematic benchmarking. Performance was assessed using Root Mean Square Error (RMSE) between estimated and ground-Truth quaternions. Results show that VQF provides stable estimates across diverse conditions with minimal tuning, XKF3i achieves high accuracy due to proprietary real-Time fusion, and Kalman-based filters perform reliably under moderate dynamics but require parameter adaptation. The study highlights the advantages of robot-guided benchmarking for filter evaluation and identifies VQF and Kalman-based approaches as particularly suited for robotics applications requiring robust and precise orientation tracking.