How a Robot can Learn to Throw a Ball into a Moving Object through Visual Demonstrations

The objective of this thesis is to extend a known probability model for Imitation Learning. Thanks to this model we can teach to a robot how to learn simple tasks, like throwing a ball into a basket, so that it can take into account initial informations related to the surroundings, i.e. that the position of the basket can change in time. We start from human demonstrations recorded with a low cost camera and we verified thanks to a robot if the model adapts well to the introduced changes