Existing datasets or built around your task.
Start with suitable licensed material or define a targeted collection for your task, camera setup and environment. Request footage with or without annotation, depending on your training pipeline.
Two ways to get started.
The route depends on urgency, permitted use, required modalities and how tightly the data must match your target task, model or embodiment.
Existing or off-the-shelf
Ask us to assess suitable footage or demonstration datasets that may already exist within qualified sources.
- Availability and rights checked before quotation
- Representative clip and schema review where possible
- Optional annotation, cleaning and repackaging
Bespoke collection
Plan a new dataset around a particular task, target environment, camera geometry or robot embodiment.
- Defined capture and task protocols
- Agreed recording, sensor and metadata requirements
- Volume, schedule and acceptance criteria scoped in advance
Tell us the specification.
A precise brief makes it easier to evaluate sources and estimate viable pilot volumes.
Learning objective
Policy imitation, perception, task reasoning, evaluation or multimodal training.
Data family
EGO mono or stereo, UMI, teleoperation, task video or other data modalities.
Recording context
Region, workplace, task, viewpoint, tool or robot embodiment.
Signals and format
Required frame rate, timestamps, sensor streams, camera calibration and delivery schema.
Volume and timing
Approximate accepted hours or episodes, desired first delivery and cadence.
Rights and QA
Permitted training uses, exclusivity, privacy constraints and acceptance checks.
Need a data quote or a pilot sample?
Tell us whether you want existing material or custom collection and we can assess the best route.