Special Issue on Robotic Vision

This is the third in a line of special issues of The International Journal of Robotics Research (IJRR) devoted to work at the intersection of computer vision and robotics. Two previous special issues on robotics and vision were published Given this is now a nearly decade-long series of issues, it is interesting to reflect on some of the changes we have seen. Perhaps the most obvious change is the growing diversity of connections between computer vision and robotics. It is now routine for robotic systems to include object or scene recognition, vision-based motion control, vision-based mapping, and dense range sensing. There are several reasons for this shift. The first is purely practical: algorithms that were computationally challenging a few years ago are now easily real-time or nearly so. However, changes in sensing have had an arguably larger impact. Even in the prior special issue, the editors noted that, '…even the simplest laser rangefinders are two to three orders of magnitude more expensive than a camera, consume orders of magnitude more power, are considerably heavier and bulkier, and capture just a slice through the scene'. This has changed drastically in just the past year with the introduction of commodity dense range sensing – high quality range and intensity images are now available in the mass market. Comparable changes have taken place in the world of algorithms. Ten years ago, SIFT features were just hitting the main stream; now SIFT and its variations are everywhere. Likewise, machine learning methods have now been leveraged to create open-source software systems with impressive scene and object recognition performance. At the same time, there is more discussion about 'vision sys-tems' at all of the robotics conferences, many of them finding inspiration in human and other biological systems. The growth of vision-based robotics can be seen in the response to this special issue. The call for papers for this issue was issued in late 2010 and a total of 44 papers were submitted. After a thorough review process, 10 papers were accepted for publication and appear in this issue. Two of these papers contain multimedia content as well. Autonomous navigation using computer vision has been a popular research theme for some time now, but it is gaining in importance due to recent natural disasters that have required serious robotic reconnaissance and rescue support. There are also obvious applications for planetary exploration. The article by Stelzer, Hirschmüller and …