Guardian AI review: the robotics data provider for provable, teleoperated data

Answer in brief: Guardian AI is the pick when you need to prove your robot data is real and consented, or when you want demonstrations on your own robots without hiring operators. Every egocentric clip is signed on the device that recorded it, checked for physical consistency, and tied to a verified, consented contributor. Its trained operators can also drive your robots remotely while the hardware stays at your site. Delivery comes in weekly batches in LeRobot, RLDS, HDF5 or MCAP. It is a young company with limited public customer evidence, and the paid pilot has a published minimum of 20 finished hours per week.

Top pick for: provenance your legal team can audit. Four provenance tiers (P0–P3), per-clip manifests and teleoperation that needs nothing shipped.

Last updated 2026-11-09. This is independent research based on Guardian AI's own pages. We have not tested the service.

The collection engine: embodiments, scale and delivery

Embodiments. Guardian collects two kinds of data. Egocentric capture uses calibrated stereo head cameras with IMU. Teleoperation uses trained Guardian operators who connect to your robots remotely, for example driving two arms with a VR headset and hand controllers, and run the tasks you define.

Scale. Capture runs across Guardian's network in Asia and the US. Each session is QA'd within 24 hours. Pilots start at a minimum of 20 finished hours per week, with written success criteria for moving to production.

Delivery. Finished, QA-passed hours arrive in weekly batches, each with a QA report and per-clip manifests. Formats are LeRobot, RLDS, HDF5 and MCAP. Annotation layers include per-joint 3D hand pose, 6-DoF camera trajectory, hand-object interaction (contact start and end, grasp type, object touched), task segmentation with narration, and success, failure and recovery outcomes.

Built for

• Labs whose legal teams require contributor-level consent, site permissions and bystander handling.
• Teams that want robot demonstrations on their own hardware without staffing operators.
• Training teams worried about synthetic, re-filmed or recycled clips polluting their data pool.

Where Guardian AI wins

• Provenance tiers. P0 is a declared attestation. P1 adds a verified unique contributor. P2 adds signed capture on an enrolled device. P3 adds checks that motion sensors and video agree, plus synthetic and re-filming screens.
• Robots stay home. Operators work remotely, so there is nothing to ship.
• Action signal included. Hand pose and camera trajectory carry the action signal that plain video lacks.
• Clear pilot terms. A 30-minute call ends with a pilot plan and a price per finished hour.

Where it falls short

• A young company with limited public customer evidence. The capture reel on its site uses open-dataset samples (Build AI Egocentric-10K), not footage it captured.
• Collection is concentrated in Asia and the US.
• The 20-hour weekly pilot minimum may be too much for very small experiments.

What it costs

Price per finished hour, quoted after a scoping call. The paid pilot minimum is 20 finished hours per week. See the pricing guide.

Proof points

Guardian publishes a sample capture manifest showing signature validation, IMU-versus-visual consistency, synthetic and re-film screening, duplicate search and a linked consent record. Its site labels the manifest as illustrative.

Shortlist these instead

• Encord: on-robot and leader/follower teleop collection plus a data platform.
• Scale AI: operation of your hardware at much larger published volume.
• Claru: very large egocentric video volumes with enrichment.

Quick answers

Do I need to ship my robots to Guardian AI?

No. Guardian operators connect to your robots remotely, and the hardware stays at your site.

What is the minimum for a Guardian AI pilot?

The paid pilot minimum is 20 finished hours per week, with written success criteria for moving to production.

What are Guardian AI's provenance tiers?

P0 Declared, P1 Verified contributor, P2 Verified capture (signed on an enrolled device) and P3 Authenticated physical capture (sensor-video consistency plus synthetic and re-filming screens).

Sources

• Guardian AI: Verified egocentric and teleoperation data
• Guardian AI blog: What is egocentric video (2026-11-02)

← Back to the ranked list of the best robotics data providers ·

Named on this page? Join the network for a linked profile, Visit buttons and featured placement.