Labelbox review: Terra as a robotics data company for coverage-driven collection

Answer in brief: Labelbox Terra is Labelbox's robotics data engine. It captures egocentric and teleoperated demonstrations on seven purpose-built, hardware-synchronized rigs, then validates and annotates them, using one episode format from pre-training through evals. Its standout feature is a diversity engine that measures coverage against your target distribution and steers collection toward gaps. Programs are scoped and quoted; no prices are public.

Directory category: Teleoperation collection · also Egocentric video. Last updated: 2026-11-09. This is independent research from Labelbox's public pages; we have not tested Terra. Figures are Labelbox's own claims.

At a glance: 1 PB+ produced as of January 2026 · 7 capture rig configurations · up to 4 synced camera streams per episode · embodiment-agnostic · pre-training, post-training and evals in one pipeline.

What Terra does for robotics data

Terra delivers data for three training stages. Pre-training: diverse egocentric and third-person video across environments, objects and tasks, auto-labeled with scene, object and action segments. Post-training: expert teleoperation and instrumented-gripper demonstrations with synchronized action streams and success labels, for imitation learning and RL fine-tuning. Evals: standardized manipulation and navigation tasks with scoring rubrics, run against your policy and reported per environment.

The seven rigs are ego mono, ego stereo (RGB plus depth), ego trio, ego plus wrist, trio plus instrumented gripper, overhead, and teleoperation (joint states, actions and video). Every rig ships to the operator pre-configured and writes the same episode format, so you can mix viewpoints in one program.

Who it fits

• Teams that care about distribution coverage across environments, objects, operators and lighting, not just raw hours.
• Labs that want pre-training data, post-training demonstrations and policy evals from one vendor in one format.
• Teams that want to commission specific tasks against a target distribution.

Strengths: an engine that knows what it has not seen

• Diversity engine: Each episode is ingested, auto-categorized by environment, object, task and operator, validated, and counted against a target. Where coverage is thin, collection is redirected there next.
• Hard quality line: Automated checks run on every episode, with human review on edge cases. Labelbox says failed episodes are re-recorded, not shipped.
• Scale: Labelbox reports more than 1 PB produced as of January 2026.
• Embodiment-agnostic: Labelbox does not build robots and works with your hardware or its own.
• Funding momentum: Labelbox announces a $100M+ Series E on the Terra page.

Honest drawbacks

• Collection runs through Labelbox's Alignerr network and specialist third-party partners, on rigs Labelbox supplies. Ask who records your data and where.
• The page says delivery comes in the formats your training stack expects, but it does not name specific formats.
• No public pricing.
• The Terra page does not describe 3D/LiDAR point-cloud annotation.

Pricing model

Custom. Most programs start with a task list and a target distribution, which Labelbox scopes before showing you the first episodes.

Notable facts

Terra data is made in homes, workplaces and Labelbox's San Francisco robotics lab, where rigs are built, operators are trained and episodes are validated. The lab is shown in a video visit by Dwarkesh on the Terra page.

Best alternatives to Labelbox Terra

• Scale AI: data factories, home collectors and in-field deployments at a stated 1,000+ hours a day.
• Encord: collection plus curation and LiDAR annotation software, deployable in your own cloud.
• Toloka: license egocentric, UMI and teleoperation datasets now, or collect self-serve.

Labelbox Terra FAQ

Does Labelbox build robots?

No. Labelbox says Terra produces the data robots learn from and the evals that measure them, and it works with your hardware or Labelbox's.

Who records Terra data?

Collection runs through the Alignerr network and specialist third-party partners, on rigs Labelbox supplies, in homes, workplaces and its San Francisco robotics lab.

Can I commission a specific task?

Yes. Most programs start with a task list and a target distribution, and the diversity engine steers collection until coverage is met.

Sources

• labelbox.com/products/terra (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.