A group of scientists from the University of Maryland and NICTA in Australia recently succeeded in creating a robot capable of learning a set of tasks on its own in order to reproduce them according to the needs of its owner.
The robot in question was thus able to modify its functionalities and gain experience in simple tasks such as cooking, simply by connecting to YouTube and watching demonstration videos.
The system in question is based on recent advances in the field of artificial neural networks, which allow artificial intelligence to become more aware of the environment in which it is located, to better identify the position and nature of the tools that 'he manipulates, to react according to the situations that present themselves to him independently.
The big step forward will have been to allow the robot to understand and associate the 2D representations of YouTube images, with elements that it knows in 3D in the real world. Something that humans do naturally, but which is more delicate for computer systems.
The development of the system allowed the robot to identify an egg and break it cleanly, when it had no knowledge of the egg in question or how to break it, before placing it in front of a set 88 videos from YouTube.
Having a robot capable of learning by itself could not only relieve the development of knowledge bases, but also to blow up the limits of artificial intelligence, the robot was then only constrained by its own envelope.