Visualized using Foxglove: DROID A large-scale Robot Manipulation Dataset (Couch)
The DROID dataset, visualized using Foxglove, is a large-scale, in-the-wild robot manipulation dataset created by many talented individuals—too many to tag!
Collected across 564 scenes and 86 tasks, the dataset consists of 76k demonstration trajectories, totaling 350 hours of interaction data. In this visualization, there are three different 3D panels showcasing the URDF of the Franka Robotics Emika Panda robotic arm’s movements. Additionally, there are image panels displaying 2D and 3D data from Zed stereo cameras, as well as 2D imagery from a wrist-mounted Zed Mini stereo camera.
The focal 3D panel uses a Turbo color map, offering several advantages for visualizing multimodal data. Turbo provides perceptual uniformity, ensuring consistent color changes across data values, and enhanced visual contrast, making subtle differences more noticeable. Unlike other color maps like Jet, Turbo avoids misleading visual cues by providing smoother transitions and better color continuity. It is also more accessible for colorblind users, improves dynamic range by clearly representing both low and high data values, and reduces visual artifacts, making it easier to detect patterns, outliers, and relationships in complex data sets.
Link to the dataset and the list of creators: https://droid-dataset.github.io/