data collection

Capture the data your robot actually needs.

Bring in teleoperation sessions, simulation rollouts, and fleet logs with every sensor synchronized and every episode described — then let agents tell you which scenarios you're still missing.

Robot learning is bottlenecked less by volume than by coverage. Agentuor's collection layer ingests raw sensor streams — RGB, depth, LiDAR, joint states, gripper telemetry, audio — aligns them on a common clock, and attaches structured episode metadata such as task, embodiment, environment, and operator. From the first upload, agents begin profiling the corpus so gaps in lighting, object variety, or manipulation type surface before an annotation budget is spent on redundant footage.

Multi-sensor ingestion

RGB-D cameras, LiDAR, IMU, joint encoders, force-torque, and audio, aligned to a shared timeline with drift correction.

Episode metadata

Task, embodiment, scene, operator, and outcome fields captured at ingest, editable later, and searchable across the whole corpus.

Sim and real, side by side

Import simulation rollouts alongside real recordings and track domain gaps between them as first-class metrics.

Coverage analytics

Agents cluster episodes by scene, object, and motion so you can see which conditions are over- or under-represented.

Collection guidance

Recommendations for the next recording session: which tasks, objects, and environments would most reduce dataset blind spots.

Governed from the start

Retention rules, access policies, and PII handling applied at upload, not retrofitted later.

how it works

Step by step

1

Connect a source

Upload from a robot, a teleop rig, a simulator, or cloud storage using the SDK or CLI.

2

Synchronize and validate

Streams are aligned, gaps and dropped frames are flagged, and episodes are segmented automatically.

3

Describe the episodes

Metadata is inferred where possible and confirmed by an operator with confidence shown.

4

Review coverage

Agents chart the corpus by condition and recommend what to collect next.

Common questions

Which robot platforms can I ingest from?

Any platform that can export timestamped sensor streams. We provide SDKs for common middleware and a generic format for everything else.

Can I import data I already have?

Yes. Existing datasets can be imported with their labels and mapped to the Agentuor schema without losing history.

Where is data stored?

In your chosen region on our cloud, or in your own storage with a hybrid deployment. See the Security page for details.

See it running on your data.

Start with a walkthrough on your own data.