Learning of feature points without additional supervision improves reinforcement learning from images

Rinu Boney, Alexander Ilin, Juho Kannala

In many control problems that include vision, optimal controls can be inferred from the location of the objects in the scene. This information can be represented using feature poin…