A package for propagating uncertainty (multivariate Gaussian, MvGaussian) through nonlinear functions and approximating uncertainty.
Includes a lightweight implementation of Kalman unscented Kalman filtering with the following features:
- Fallback to linear methods for transition if the state predictor is linear
- Fallback to linear methods for observations if the state observer is linear
- Consistent use of the square-root form for improved numerical stability
- Automatic observation space reduction to remove non-finite observations
- Optinal multithreading if state transition or observation is computationally intensive
- Optional outlier suppression scheme for observations