Physics-informed state observer for unknown linear autonomous systems with noisy measurements

Abstract

State estimation is a pivotal element in navigation tasks of autonomous vehicles. This technique is mainly applied when either a required measurement is not available or when the amount of available sensors in the platform are limited. Most of the state estimation algorithms available in on-board control modules use kinematic models as prior model to estimate the states of the autonomous system. However, these simple kinematic models do not consider dynamic terms and physical properties which can lead to biased state estimates. To overcome this issue, this paper proposes a physics-informed state observer for unknown linear systems under partial and noisy measurements. The proposed approach fuses two complementary concepts for state estimation and dynamics identification. The proposed approach is capable to obtain reliable state estimates whilst attenuating the level of noise. Lyapunov stability is used to derive an appropriate update law for the construction of physics-informed estimate model. Simulation studies are given to show the advantages and challenges of the proposed approach.