Data
Data and simulation resources to train and test robots.
- Objaverse-XLOpen source
Ten million 3D objects for populating simulators and training vision.
- KITTIOpen source
The benchmark that defined visual odometry and driving perception.
- M2DGROpen source
Multi-sensor ground robot sequences indoors and outdoors with ground truth.
- GraspNet-1Billion
Large-scale benchmark and baseline for general object grasping.
- Ego4DOpen source
3,600 hours of egocentric human video used to pretrain manipulation models.
- OmniObject3D
6,000 real scanned objects with meshes, videos and point clouds.
- TartanAirOpen source
Photorealistic simulated SLAM sequences with hard weather and lighting.
- EPIC-KITCHENS-100
100 hours of first-person kitchen video with action and object labels.
- ACRONYMOpen source
17.7 million simulated parallel-jaw grasps on 8,872 ShapeNet objects.
- YCB object setOpen source
The standard benchmark objects for manipulation, ready for simulation.
- EGAD
Evolved 3D-printable objects that span grasp difficulty and shape complexity.