LUMBER Dataset 

Occluded Video Dataset of a Humanoid Robot 

The LUMBER (Looking Upon a Moving Bipedal Robot) dataset contains 15-sec sequences of the humanoid robot, Digit. With 110 videos, there are 49,152 images, depth images, and annotations. Our dataset required no external fiducial or motion capture markers. It also contains a variety of obstacles including fast, whole-body occlusions, static partial occlusions, and materials similar to Digit's parts. 

Heavy and Full-Body Occlusions

Partially-Observable Occlusions

RGB-D Data and Annotations

We provide 1280x720px RGB and depth images. For each frame, our dataset includes each link's 6DoF pose, joint values, and segmentations. 

Preprint coming soon!