1: \begin{abstract}

2:     Measurement outliers are unavoidable when solving real-world robot state

3:     estimation problems.  A large family of robust loss functions (RLFs) exists

4:     to mitigate the effects of outliers, including newly developed adaptive

5:     methods that do not require parameter tuning.  All of these methods assume

6:     that residuals follow a zero-mean Gaussian-like distribution.  However, in

7:     multivariate problems the residual is often defined as a norm, and norms

8:     follow a Chi-like distribution with a non-zero mode value.  This produces a

9:     ``mode gap'' that impacts the convergence rate and accuracy of existing

10:     RLFs.  The proposed approach, ``Adaptive MB,'' accounts for this gap by

11:     first estimating the mode of the residuals using an adaptive Chi-like

12:     distribution.  Applying an existing adaptive weighting scheme only to

13:     residuals greater than the mode leads to more robust performance and faster

14:     convergence times in two fundamental state estimation problems, point cloud

15:     alignment and pose averaging. 

16: \end{abstract}

17: