Robotic-Arm-Based Validation of Orientation Estimation Filters for Gravity Artifact Removal in 6DoF Sensors

This study investigates whether orientation estimation filters can reliably predict sensor orientation to enable the removal of gravitational components from acceleration signals. Data were collected using a custom-fused sensor system integrating accelerometer and gyroscope measurements. Three orientation filters - Madgwick, Extended Kalman, and Complementary - were applied to estimate sensor orientation. For validation, a robotic arm was used as ground truth reference. To avoid singularities associated with Euler angle representations (gimbal lock), each axis was compared to the gravity vector to compute an angle-to-gravity metric. Results indicate that all three filters successfully estimate sensor orientation, with the Madgwick filter achieving the best overall accuracy while maintaining high computational efficiency. This supports its suitability for removing gravitational artifacts from accelerometer data, as illustrated by an example of a pure translational signal.