Centroidal Voronoi Tessellation MAP-Elites

CVT-MAP-Elites module for Sferes2.

Python version

This is the C++ version. It will be merged soon in sferes2. A python version of CVT-MAP-Elites is available here: https://gitlab.inria.fr/resibots/public/py_map-elites

Authors:

• Vassilis Vassiliades (vassilis.vassiliades@inria.fr)

Description

Centroidal Voronoi Tessellation MAP-Elites is a natural extension of MAP-Elites to handle high dimensional behavior spaces. The number of niches in MAP-Elites is dependent on the dimensionality of the problem and the discretizations used. In CVT-MAP-Elites, the number of niches is set before-hand. The location of these niches is determined in a way that the behavior space is partitioned into equisized regions. This is done using a centroidal Voronoi tessellation (https://en.wikipedia.org/wiki/Centroidal_Voronoi_tessellation). New points fall into the region of the closest centroid.

Example on Rastrigin function:

• MAP-Elites

• CVT-MAP-Elites