Adaptive Kalman Filtering for Unknown External Measurement Noise

In practice, sensor measurements can often be subject to environmental noise, which can degrade the accuracy of state estimators. Traditional Kalman filters typically assume that sensor noise is Gaussian, which can reduce their effectiveness and lead to poor performance when sensors are affected by unknown external noises that might be nonzero-mean and non-Gaussian. This letter proposes two adaptive Kalman filters (AKFs) that can compensate for unknown, potentially nonzero-mean external noise, an aspect often overlooked in the literature. Compared to variational Bayesian Kalman filters, our proposed AKFs achieve more accurate state estimates with lower computational complexity, without needing assumptions about the probability distribution of the noise.