In this paper, we present a visual simultaneouslocalization and mapping (SLAM) system for humanoid robots.We introduce a new binary descriptor called DLab that exploitsthe combined information of color, depth, and intensity toachieve robustness with respect to uniqueness, reproducibility,and stability. We use DLab within ORB-SLAM, where wereplaced the place recognition module with a modification ofFAB-MAP that works with newly built codebooks using ourbinary descriptor. In experiments carried out in simulation andwith a real Nao humanoid equipped with an RGB-D camera, weshow that DLab has a superior performance in comparison toother descriptors. The application to feature tracking and placerecognition reveal that the new descriptor is able to reliablytrack features even in sequences with seriously blurred imagesand that it has a higher percentage of correctly identified similarimages. As a result, our new visual SLAM system has a lowerabsolute trajectory error in comparison to ORB-SLAM and isable to accurately track the robot’s trajectory.