Unscented

Unscented is an adjective most often encountered in estimation theory and signal processing, referring to methods that use the unscented transform to manage nonlinear uncertainty. The core concepts appear in the Unscented Transform and its applications, notably the Unscented Kalman Filter (UKF) and related algorithms.

The Unscented Transform represents a probability distribution, typically Gaussian, with a small set of deterministically chosen points called sigma points. These points are selected to capture the distribution’s mean and covariance. Each sigma point is propagated through the nonlinear state transition and measurement functions. The transformed points are then recombined to yield an approximate distribution of the state, including a new mean and covariance. This approach avoids linearizing the nonlinear functions and tends to preserve more information about the true distribution.

In filtering, the UKF uses the unscented transform to perform prediction and update steps. It often provides more accurate estimates than the extended Kalman filter for strongly nonlinear systems and does not require calculation of Jacobians. Variants such as the Scaled Unscented Transform (SUT) and the Scaled Unscented Kalman Filter (SUKF) introduce scaling parameters to better capture higher-order moments and improve robustness.

Applications span robotics, navigation, aerospace, and other domains where nonlinear state estimation is essential. Limitations include computational cost relative to linear filters and the need to assume approximately Gaussian process and measurement noise, with performance diminishing as system dimensionality grows.

Note: outside the technical sense, unscented also serves as a general English term meaning without scent.