RNN-based State and Parameter Estimation for Sparse Magnetometer-free Inertial Motion Tracking

Abstract

Inertial measurement units are widely used for motion tracking in numerous applications. Combining magnetometer- free sensor fusion with sparse sensor setups yields advantages in cost, effort, and usability but makes observer design a challenging task. Recently, a data-driven solution that utilizes recurrent neural networks as observers (RNNOs) has been developed. There, it is assumed that the geometry of the kinematic chain (lengths of segments, sensor-to-segment positions) is known. However, in practice the geometry is often unknown. Here, we show that RNNO can estimate the relative pose of a three segment kinematic chain with double hinge joints while (implicitly) identifying its geometry.