MARG Attitude Estimation Using Gradient-Descent Linear Kalman Filter

The magnetic, angular rate, and gravity (MARG) sensor array has been widely used for attitude estimation tasks. In this article, we report our new advances on related fusion algorithm based on the gradient-descent algorithm (GDA). Combining with complementary filters, GDA has been very popular for attitude estimation in industrial applications. The integration of the Kalman filter introduces covariance information and significantly improves in-run quality control. Some useful results are derived to build up the framework of a novel linear Kalman filter called GDA-LKF. A new simplified linear measurement quaternion model is proposed. This article also deals with the analytical adaptive problem of determining gradient-descent step length. We, for the first time, give the solution in the sense of least square based on aided vector measurements. Simulations are carried out to validate the noise sensitivity and covariance characteristics. The proposed schemes are also evaluated by real-world experiments that provide the audience with comparisons on accuracy, convergence, and execution time consumption performances between the proposed GDA-LKFs and recent representative methods. The results show that the proposed GDA-LKF can accurately estimate attitude with fast convergence and high accuracy. Note to Practitioners—Attitude estimation using magnetic, angular rate, and gravity (MARG) sensors is very common for ground, air, and underwater vehicles. The novel findings in this article aim to give the engineers a brand new perspective on the related filter design according to the widely employed gradient-descent algorithm (GDA) method. Some specific techniques, e.g., optimal adaptive law of the designed scheme, will also bring flexibility to the implementation with multiple considerations of object motions.