Process Noise Adaptive Kalman Filtering with Stein Variational Gradient Descent

Adaptive Kalman filtering has become a widely used state estimation method when the system noise statistics are not available a priori. In this article, we consider the application of Stein variational gradient descent (SVGD) in adaptive Kalman filtering for the linear Gaussian state-space model with inaccurate process noise statistics. The proposed algorithm separately estimates the system state and noise parameters with the Kalman filter and SVGD online inference. Compared to the existing state-of-the-art adaptive Kalman filters, the proposed method can approximate the noise statistics better and thus provides higher state estimation accuracy, which is confirmed by the simulation results on a target tracking example.