Key points are not available for this paper at this time.
For autonomous robots, the ability to classify their local surroundings into traversable and non-traversable areas is crucial for navigation. In this paper, we address the problem of online traversability analysis for robots that are only equipped with a Kinect-style sensor. Our approach processes the depth data at 10 fps-25 fps on a standard notebook computer without using the GPU and allows for robustly identifying the areas in front of the sensor that are safe for navigation. The component presented here is one of the building blocks of the EU project ROVINA that aims at the exploration and digital preservation of hazardous archeological sites with mobile robots. Real world evaluations have been conducted in controlled lab environments, in an outdoor scene, as well as in a real, partially unexplored, and roughly 1700 year old Roman catacomb.
Bogoslavskyi et al. (Sun,) studied this question.