Multi-Sampling-Based Distribution Simulation for Nonlinear Measurement Error Mitigation

In this study, we propose a technique to mitigate estimation errors caused by nonlinear measurements. The proposed technique considers the well-known measurement noise variance and innovation as individual samples to form their respective distributions. The calculated distribution is changed to a normalized weight to compensate for the observed measurement error, and a pseudo-measurement is calculated and used as the measurement update. Additionally, the noise variance of the pseudo-measurement is newly calculated using likelihood optimization. As a result, it was confirmed that when using the proposed technique, stable performance is achieved even if non-linear measurement errors exist for a certain period of time. In particular, it was confirmed that the proposed technique converges to the true value faster than existing filters when nonlinear errors exist in measured values and return to normal. The proposed method can construct not much change to the conventional filter as only uses predicted value and observed measurement.