No episode media available here
This is a representative still, not playable footage. Episode video for this dataset lives with its original source.
QT-Opt Kuka Grasping
Closed-loop vision-based grasping
Google DeepMind Robotics
RobotData indexes this dataset's metadata. Attribution and licensing stay with the original publisher.
580,000 real grasp attempts collected over four months across seven Kuka IIWA arms for QT-Opt, a distributed off-policy Q-learning system that reached 96% grasp success on unseen objects. The canonical demonstration that scaled self-supervised robot data beats hand-scripted grasping.
Reported structure
Values below come from the dataset's own documentation. Missing fields are shown as not provided rather than estimated.
1 cameras
7 DoF
10 Hz
96% success
Data modalities
Episode explorer
No episode media has been added for this dataset, so there is nothing to play. The explorer appears as soon as real clips exist — either uploaded here or extracted from the source.
Contribute episodesLinked models, robots & papers
Resolving related models, datasets and papers…
Related datasets
Real-world IndexedOpen X-Embodiment
Source: Open X-Embodiment Collaboration
- 1.0M
- 5.6K
- 3
Real-world IndexedBridgeData V2
Source: UC Berkeley RAIL
- 60K
- 333.9
- 4
Real-world IndexedLeRobot Community SO-100 Datasets
Source: Hugging Face LeRobot
- 23K
- 481
- 2
Real-world IndexedDROID
Source: Stanford IRIS & REAL Labs
- 76K
- 350
- 3