Learning in robots

Robot learning is a research field that brings together all attempts to endow robots with learning capabilities. A robot is a mechanical system that perceives its environment with sensors and that acts in this environment with actuators, so that the action is adapted to the perceived situation. As a result, robot learning algorithms have to deal with a limited life-span, expensive failures, high dimensionality, and real-time requirements. Furthermore, the robot is required to choose what actions to take. These requirements pose both hard constraints and opportunities on the learning processes and methods that can be used.

Robot learning is a broad field that does not yet benefit from clearly defined theoretical approaches. However, it largely profits from the theoretical background in many areas. The rigorous treatment of learning methods in the data mining and machine learning community allows adapting general machine...