Adaptive Kalman Filter via Just-in-Time Learning for Robots with Unknown Dynamics

In many practice control processes, the measured signal is often noisy, which requires some methods to estimate the actual signal despite the noise. Kalman filter is a popular and effective information fusion state estimation method in many application fields. Conventional Kalman filtering mainly focuses on estimating the state of a known system. The mathematical model in this method is often challenging to obtain as the system becomes increasingly complicated. Aiming at the unknown system, this paper studies a Kalman filtering method based on just-in-time learning. The proposed method can adaptively obtain the actual model of the unknown nonlinear system through the process data and then realize the information fusion with the measurements under the framework of Kalman filtering to accurately estimate the state. The state estimation experiment in the ground vehicle lateral control system verifies the effectiveness of the algorithm.