Manage outliers in pose distribution
Opened this issue · 0 comments
hidmic commented
Feature description
Existing estimators in Beluga for distribution mean and covariance include unweighted and weighted maximum likelihood estimators (i.e. sampling mean and covariance). These can be swayed by outlier in the distribution, that may not always be strongly mono-modal. Other techniques exist: we know Nav2 AMCL uses a form of clustering, of which there are many variations, and robust estimation of location and dispersion is on its own an entire field in statistics. We have to explore these techniques.