In this work, we present a three-part system that automatically sorts books on a shelf using the PR- 2 platform. The paper describes a methodology to sufficiently detect and recognize books using a multistep vision pipeline based on deep learning models as well as conventional computer vision. Furthermore, the difficulties of relocating books using a bi-manual robot along with solutions based on MoveIt and BioIK are being addressed. Experiments show that the performance is overall good enough to repeatedly sort three books on a shelf. Nevertheless, further improvements are being discussed, potentially leading to a more robust book recognition and more versatile manipulation techniques.