Robotic grasping in agile production

Recent developments in robotics and deep learning have enabled high-level robotic tasks to be learned from simulated or real data. In this chapter, the task of robot grasping is covered, where a robot manipulator learns a grasping model from perceptual data, such as RGB-D or point clouds. The chapter is presented in context of robotics for agile production, thereby providing requirements and limitations that are relevant for deep learning in robotics. An overview of different approaches is given with special attention to the evaluation of robotic object grasping and the potential follow-step of object manipulation. In addition, a list of data sets is provided that utilize simulation to generate training data for object grasping.