On clustering for pose estimation

[hidmic]

Connected to #85. Belief tracking with particle filters may run into a multi-modal distribution during operation. beluga currently supports weighted mean and covariance statistics for pose estimation, and so it exhibits biased means (w.r.t. to the particles that are actually tracked) and inflated covariances in such conditions. nav2_amcl, on the other hand, works with clusters to reduce mean bias but also artificially reduces covariance.

Should beluga_amcl support nav2_amcl like pose estimation for feature parity? What would be the correct approach to pose estimation in the face of multi-modal distributions?

[glpuga]

I agree this needs to be on the roadmap, but I don't think it's necessarily high priority between now and June.

[nahueespinosa]

Closing the discussion as we're already working on this: #275.