Using Linear Splines to Continuously Develop Understanding of Affordances

Understanding affordance is a bottom-up process that lets simple visual features be used as clues on how to use an object, so that reasonable interactions with objects can be achieved, without any semantic knowledge. Linear splines are proposed as a way for robots to fast and continuously learn how their actions affect their surroundings. The general conclusions that they can draw from only little experience, using this method, are seen as an example of understanding affordances, and the specific case investigated by the robot created as a part of this thesis, is how pushing actions affect small blocks. This information will be used by the robot to perform goal directed tasks, namely to push the block to specific predetermined positions. A way to make generalised predictions, so that actions on new shapes can be performed in a reasonable way without any previous interaction, is proposed. It becomes obvious that good inner representations of actions and reactions are necessary, as well as some management of learned relations