Object localization in range data for robotic bin picking

This paper describes an approach to solve the bin picking problem. In many industrial processes, product parts, which have to be assembled, are delivered scrambled in boxes. Usually these parts have to be picked out of the box manually to feed them into an automated process. Using an industrial robot for this task is very difficult. This problem is not solved in general up to now. Our flexible approach uses knowledge about the form of the objects to find them in range data. We compare the 2.5 D-appearance of simulated object poses with the real range data in two different steps, and find the best matching pose of the object. This approach can handle many different kinds of objects and takes features of range sensors into consideration to improve the accuracy and robustness of the object localization.