Localization: Smoother pose

[jaagut]

Is your feature request related to a problem? Please describe.

Currently, the localization produces very bouncy pose results.
A large reason for this is the clustering method.

Describe the solution you'd like

Improve the parameters and clustering (Gaussian mixture model)

Describe alternatives you've considered

[Flova]

Clustering it would make it more bouncy. These are two issues. One is due to the missing clustering the estimate can be in between two clusters and the other one is a noisy output. Clustering can lead to the estimate jumping between two better matching solutions compared to the current approach. The observed noise comes from a large dependency on the vision estimates that are noisy due to the projections. If we had a better odometry estimate the estimate would be smoother. We could implement a filter afterwards but this adds yet another layer.

[Flova]

Closing due to #361