AMCL pf_resample_limit returns max_samples for k=1

[christofschroeter]

The function pf_resample_limit implements the adaptive particle number for the particle filter, following Dieter Fox' paper "KLD-Sampling: Adaptive Particle Filters".

It takes as input a number k of supported bins, and computes the number of required particles, basically following eq. (7) from the paper. However, the equation cannot calculate a valid value for k=1 (because of division by k-1).
In the code, the case k=1 is handled specifically, returning max_samples.

I don't think that is a good handling of that case. For k>1, the calculated particle limit increases with k. k=1 actually means the lowest possible uncertainty estimate. It doesn't seem to make any sense to use the lowest number of particles for k=2 supported bins, but when the number of supported bins is further reduced to 1, suddenly the filter increases the number of particles to max.

Since the formula cannot handle a value of k=1, my suggestion would be to calculate with k=2 when the input parameter k is given as 1. That would seem like a slight overestimation of uncertainty, but much less than just returning max_samples.