CMA-ES dependance on sigma #4
Description
I was thinking of a possible experiment to illustrate the importance of sigma. If we have a bunch of locally optimal solutions calculated with DNJ, each of them with their corresponding times/consumptions (depending on the problem). We could see whether CMA-Bézier --while starting from the same initial seed route (the \mu of the multinormal), but with different sigmas, ends up finding one local minimum or another.
In addition, it would perhaps be illustrative to show the whole cloud of trajectories in the population as light gray curves, and depicting in red the best curve in the population. I mean already in configuration space.
Then we can make a gif showing not only the evolution of the optimal solution but also giving some information on the distribution.
Might be messy, but if trajectories are very light gray and the whole visualization is perceived like a "cloud of trajectories", I think it might be cool to show.