Menglong Zhu at Penn has given PR2 a fantastic new skill: the ability to read. Using the literate_pr2 software he wrote, PR2 can drive around and read aloud the signs that it sees. Whether it's writing on a whiteboard, nameplates on a door, or posters advertising events, the ability to recognize text in the real world is an important skill for robots.
Although performing OCR on text is not a new technology, performing it in the real world is much more difficult as text can be written anywhere and first you have to find it. Menglong's code is able to detect areas of text in a camera image and perform text recognition on them separately.
This is another great contribution to ROS from the GRASP Lab. For more information, please see the literate_pr2 page on the ROS wiki.