Robots play an important role in the automation of manufacturing processes. Specifically the automotive and electronics industry profited from the high speed and repeatability of robotic arms. However, the presence of robots in industries dealing with deformable objects, such as the textile and food industry, is less pronounced. This is due to the high amount of deformations that occur on soft objects. In order to reach new productivity gains in these industries, we need robotic manipulators that can handle this large configuration space. An answer can be found in the domain of machine learning where a solution is learned instead of programmed. In particular, the use of deep reinforcement learning has recently solved very complex problems such as winning Atari games and controlling a robotic arm without prior knowledge and using visual feedback only. Therefore, we will in this project investigate the use of deep reinforcement learning to train a robot controller to fold clothing based on visual human task demonstrations. This solution can
relieve employees from repetitious tasks and let them focus on value-added activities requiring a human operator.