Firsthand review: the robotics data company that publishes its sync error

Answer in brief: Firsthand collects custom egocentric video datasets for embodied AI and robotics, to your spec. Its instrumented rig ships ten synchronized streams per episode: a 4K head camera, two wrist cameras, depth, IMU, 3D hand and body pose, segmentation and action labels. Firsthand reports a median cross-stream sync error of 1.12 ms. Licenses are buyer-owned and perpetual, and a free sample pack needs no call or NDA. Billing is per validated hour, at a rate quoted on request.

Directory category: Egocentric video. Last updated: 2026-11-09. This is independent research from Firsthand's public site; we have not downloaded or tested its data. Figures are Firsthand's own.

At a glance: 10,585 validated hours delivered · 1.12 ms median sync (p99 1.94 ms) · median 48 hours to first delivery · RLDS, LeRobot, WebDataset, HDF5, zarr, Rerun .rrd · 100% documented consent.

What Firsthand does for robotics data

You write the spec: environments, skills, conditions and hours. Firsthand recruits and trains contributors, runs QA on every batch, and delivers footage with depth, hand and body pose, metadata and a documented consent chain. Its instrumented rig hardware-triggers every sensor from a single PTP grandmaster clock. Calibration is re-solved at the start and end of each session, and drift is recorded in metadata. The action taxonomy covers 97 verbs and 512 nouns, labeled to the frame and human-reviewed.

Who it fits

• Robot foundation-model teams pre-training on rich first-person human data.
• Buyers who want to inspect real data and schema before talking to sales.
• Teams that need clean commercial rights, with exclusivity available.

Strengths: transparency as a feature

• Measured sync: 1.12 ms median and 1.94 ms p99 across 12,480 measured frames. Frames above the ceiling are rejected and recaptured.
• Published coverage gaps: Firsthand publishes a coverage matrix showing where its data is deep and where it is thin.
• Licensing: Commercial, buyer-owned, perpetual and irrevocable, with exclusivity available.
• Speed: A median 48 hours from signed spec to first validated batch, in environments where coverage is already deep.
• Global reach: Firsthand describes an engagement, which it labels illustrative and anonymized, that delivered 5,000 videos across 16 countries in 19 days. Custom collection is offered across 150+ countries.
• Try first: A 2.4 GB pack with 40 episodes, 4 environments and 8 synced streams; the download link is sent by email.

Honest drawbacks

• Human egocentric data only, with no robot teleoperation trajectories, so actions need retargeting before a policy can learn them.
• A second capture tier uses contributor-worn 360 cameras with no depth, no wrist views and no sub-millisecond sync. The published sample clips come from this tier. Filter by tier in the metadata.
• The global case study is anonymized at the buyer's request.
• Per-hour rates are not published.

Pricing model

Billed per validated hour: hours that passed sync and QA and shipped. Recaptured and rejected hours are excluded. The rate is given on request.

Best alternatives to Firsthand

• Toloka: 25,000+ hours of off-the-shelf egocentric household video.
• Knoxlabs: US in-home egocentric and glove capture.
• Labelbox Terra: egocentric plus teleoperation in one episode format.

Firsthand FAQ

How is Firsthand data priced?

Per validated hour, meaning hours that passed sync and QA and shipped to a buyer. The rate is not published.

Can I see Firsthand data before talking to sales?

Yes. Sample clips are viewable on its site, and a free 2.4 GB pack is available without a call or NDA. The download link is sent by email.

Does Firsthand provide robot teleoperation data?

No. Firsthand collects human egocentric data, not robot teleoperation trajectories.

Sources

• ifirsthand.com (read 2026-11-09)

← Back to the robotics data companies directory ·

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