DROID labels
Scheduled, license-clearedBroad multi-scene manipulation with synchronized views.
- Source size
- 76k trajectories (official release)
- License
- CC-BY 4.0, cited to the official release
Physical-event labels for robot datasets.
Haptica adds five physical-state labels to datasets you already use. An ID-keyed overlay joins the labels to source episodes. The source video stays with the original dataset.
What you receive
- A per-timestep physical-state sequence
- A timestamped physical-event record
- Source episode IDs
- Corpus-specific confidence and abstention
- Held-out quality and coverage
- Grouped train and evaluation splits
- Provenance
- Corrections history
- A training and evaluation guide
Confidence is calibrated and checked on held-out data from that dataset.
Availability
- Available
- Labels complete, checked, and cleared for delivery.
- In production
- License-cleared. Labeling underway.
- Scheduled, license-cleared
- Cleared and queued for labeling.
- Free evaluation set
- Free labels for evaluation. Never sold.
Datasets
BridgeData V2 labels
Scheduled, license-clearedThe corpus many teams already fine-tune on. Labels at the temporal resolution supported by the source data.
- Source size
- 60,096 trajectories (official release)
- License
- CC-BY 4.0, cited to the official release
RoboMIND labels
Scheduled, license-clearedMulti-embodiment coverage.
- Source size
- 107k trajectories (official release)
- License
- Apache-2.0
Instrument-grounded labels
Labels for instrumented datasets can be checked against the dataset's own force-torque traces.
Delivery
Each delivery includes a manifest, checksums, provenance, a corrections history, and an integration call.