Visualized using Foxglove: The Hugging Face 🤗 Le Robot Unitree H1 Datasets
The Unitree H1 folding clothes and arranging items datasets, captures a robot performing common daily tasks, offering a rich source of data for advancing robotics, computer vision, and AI research. It includes 38 episodes with a total of 19,000 frames recorded at 50 fps, featuring both 19-dimensional state vectors and stereoscopic RGB images at 1280x720 resolution. Each frame also includes a 40-dimensional motor command vector, enabling precise action tracking.
Designed for efficiency, the dataset is stored in Parquet format, making large-scale data processing seamless. We converted it to MCAP for seamless visualization using Foxglove. The visualization no only displays the RGB images but also plots the joint states, adds the Unitree H1 URDF to a 3D scene, and incorporates DepthAnythingV2 in the middle colored set of depth images.
Visualize the dataset directly in Foxglove for your self. Links to the dataset and DepthAnythingV2 project in the comments 👇