1: \begin{abstract}

2: We present a Kalman smoothing framework based on 

3: modeling errors using the heavy tailed Student's t distribution,

4: along with algorithms, convergence theory, open-source general implementation, 

5: and several important applications. 

6: The computational effort per iteration grows linearly with the length of the time series, 

7: and all smoothers allow nonlinear process and measurement models. 

8: 

9: Robust smoothers form an important subclass of smoothers within this framework.  

10: These smoothers work in situations where measurements are highly contaminated

11: by noise or include data unexplained by the forward model.  

12: Highly robust smoothers are developed by modeling {\it measurement} 

13: errors using the Student's t distribution, 

14: and outperform the recently proposed

15: $\ell_1$-Laplace smoother in extreme situations

16: with data containing 20\% or more outliers.

17: 

18: A second special application we consider in detail

19: allows tracking sudden changes in the state. 

20: It is developed by modeling {\it process} noise using 

21: the Student's t distribution, and the resulting smoother 

22: can track sudden changes in the state.

23: 

24: These features can be used separately or in tandem, 

25: and we present a general smoother algorithm and open source implementation, 

26: together with convergence analysis that covers a wide range of smoothers. 

27: A key ingredient of our approach is a technique to deal with the 

28: non-convexity of the Student's t loss function.  

29: Numerical results for linear and nonlinear models illustrate the performance

30: of the new smoothers for robust and tracking applications, 

31: as well as for mixed problems that have both types of features. 

32: 

33: 

34: \end{abstract}

35: