Claru vs XDOF: Match the Data to Your Training Objective
Last updated: November 11, 2026. If anything here is inaccurate, email [email protected].
Human coverage and robot grounding are different decisions
- Evaluate Claru when you need more human household/workplace activity with a defined task distribution.
- Evaluate XDOF when robot interaction, embodiment and action trajectories are central to the purchase.
- XDOF is not teleop-only: its own role description includes egocentric capture and UMI-style grippers.
- Do not choose on hours alone. Test channel completeness, alignment and downstream compatibility on a representative sample.
Compare the supervision, not a false human-versus-robot binary
XDOF's public ABC-130K release gives buyers a concrete robot-demonstration reference: more than 130,000 episodes across 195 bimanual tasks, captured on stations with I2RT YAM arms. Its broader data work includes teleoperation, egocentric capture and UMI-style grippers. Release details · Broader modalities
Claru's household program documents first-person chest/waist video at humanoid operating height, task quotas and acceptance gates. The workplace case provides separate evidence of commercial activity collection. That supports evaluating human-activity coverage, not claiming the video is already a robot-action trajectory. Household case · Workplace case
A training plan can use both. Human activity can expand the behaviors and settings observed, while a robot-demonstration purchase can address specific embodiment/action requirements. Whether the combination improves your model is a pilot question; neither these program counts nor a vendor's positioning prove it.
Why a buyer would evaluate XDOF
A concrete robot release to inspect
ABC-130K documents the task count and station hardware. Inspect that release's schema, license and task mix separately from any custom quote. Source
Multiple ways to capture interaction
XDOF explicitly describes real-hardware teleop, egocentric capture and UMI-style grippers, plus validation of camera, joint and gripper streams. Source
Systems and tooling
Its company offer includes production datasets, advanced robotic systems and tooling. Scope what your team needs beyond the files themselves. Source
Why a buyer would evaluate Claru
Household tasks at operating height
The household program uses chest/waist-mounted cameras and a versioned task taxonomy. Check that camera height, activity framing and task completeness match your next requirement. Source
Quota-driven acceptance
The household case records task/country quotas and acceptance gates. Specify what will count as accepted output before commissioning more capture. Source
Workplace distribution
Claru's commercial case covers real workplace settings as a separate program. Inspect those environments rather than treating household hours as evidence of industrial coverage. Source
Claru vs XDOF: the sample acceptance checklist
| Dimension | XDOF | Claru |
|---|---|---|
| Documented data reference | ABC-130K bimanual manipulation episodes. Source | Household first-person tasks and separate workplace footage. Household · Workplace |
| Capture evidence | Teleop, egocentric and UMI-style collection. Source | Household chest/waist video at humanoid operating height. Source |
| Quality/process evidence | Data-engineering role describes temporal alignment and validation across camera/joint/gripper streams. Source | Household case describes task match, stability, full-activity review and task quotas. Source |
| Buyer must validate | Required embodiment, action/state fields, timestamps, calibration and loader compatibility. | Activity distribution, camera framing, complete tasks and any separately required action/sensor channels. |
Share target tasks, environments, camera constraints and acceptance rules.
Request a DatasetEvaluate XDOF when…
- The training recipe requires robot-specific observations and interaction/action data.
- You need to evaluate its robot systems, capture modalities or tooling alongside data.
Evaluate Claru when…
- You need to expand real human-activity coverage across household or workplace tasks.
- Your collection needs a clear task taxonomy, quotas and accepted-video rules.
Run a two-provider technical pilot
Do not make a one-for-one substitution until these checks pass for your own training recipe.
Freeze the loader contract
Define observations, actions, states, units, frequency and coordinate frames. Mark which channels must be recorded and which can be derived.
Inspect the complete execution
Load a task from start to finish. Check camera placement, timestamp alignment, missing segments and any embodiment mismatch. Record rejection reasons against both samples.
Measure the intended use
Run your downstream evaluation with the same task split. Decide whether human video, robot demonstrations or a mixture is justified. Then scope accepted output, rights and annotation.
Claru's human-activity program record
Inspect each part of the data requirement
Expand the shortlist or review the underlying collection evidence.
Claru vs XDOF: technical buyer questions
Can Claru video replace XDOF robot demonstrations?
Not automatically. If your model needs synchronized robot states and actions, verify those channels in the proposed delivery. Human activity video and robot trajectories address different supervision requirements; a matched pilot determines whether either or both fit.
Is XDOF restricted to teleoperation?
No. Its public robotics-data role lists real-hardware teleoperation, egocentric capture and UMI-style grippers. This comparison does not describe XDOF as teleop-only. Source
Does ABC-130K cover every commercial XDOF offering?
No. ABC-130K is a specific release, not a complete commercial catalog or custom contract. Confirm the intended dataset's tasks, embodiment, license and schema directly.
Which is cheaper for my model?
A comparable public numeric buyer rate was not established from the offering pages reviewed November 11, 2026. Compare the actual cost of accepted, compatible data including annotation, rights and any hardware/tooling work, not unrelated hour or episode totals.
Define the human-activity dataset you need
Share target tasks, environments, camera constraints and acceptance rules.