Dataset · Salesforce
Salesforce robot trajectories with 3D scene flow
Salesforce/3d_optical_flow_droid
Processed DROID robotics dataset with optical and 3D scene flow annotations for training robot control models.
What's inside
The dataset contains about 33 thousand robot operation trajectories from 15 laboratories (total volume from 10M to 100M records), augmented with optical flow and 3D scene flow annotations. Data is organized into individual tar.gz archives: each contains JSON metadata, robot states and actions in HDF5 format (trajectory.h5), and RGB images from left and right cameras.
What you can do with it
- Training multimodal models and robot manipulation policies (VLA)
- Training 3D motion estimation (scene flow) and optical flow algorithms in real scenes
- Tasks
- Robotics
- Size
- 10M–100M rows
- Downloads this month
- 21.7K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("Salesforce/3d_optical_flow_droid")