A Two-Stage Multiplicative Kalman Filter for Attitude Estimation of the Human Wrist
Daniel S. Eliahu, Gabriel H. Elkaim and Renwick E. Curry, University of California, Santa Cruz
This paper develops a high-performance filter for attitude estimation of the human wrist. It leverages state-of-the-art estimation practices developed by researchers in the aerospace industry while incorporating several domain specific improvements. One potential use of the filter is for the improved diagnosis of movement disorders such as Parkinson's Disease or Essential Tremor. Due to the dynamic nature of these diseases the filter is designed to handle very large rotation rates. The algorithm fuses data from low-cost gyroscopes and accelerometers. It tracks attitude using quaternion error states and is able to dynamically track the gyroscope bias - allowing the filter to be used in a commercial application without calibration. The filter is tested using a MATLAB simulation as well as with a mechanical test rig. During testing with the mechanical rig the rotation rates of the limb exceed 260 degrees per second and the filter is able to maintain an RMS error of less than 1 degree.