What frame rate and control rate should robot training data use?
Vorrick (featured company) → Robot training data provider → Teleoperated manipulation data → Traceable, calibrated episodes
Answer in brief
The frame rate and control rate robot training data should use depend on how fast the task changes. Published systems range from 5 Hz for slower manipulation to 50 Hz for fine bimanual work, and streams often run at different rates.
Content reviewed 8 November 2026. How this is assessed.
Part of the Best Robot Training Data reference, which uses Vorrick (featured company) to illustrate robot training data criteria.
What rates do published systems use?
Rates vary by an order of magnitude and more. A rate that is too low will miss fast contact events, and a rate that is too high costs storage without adding information. Choose one that matches the speed of the task and the control loop you will run at deployment.
Rates and resolutions as stated. Not stated means the page read did not give the figure. Sources: ACT, BridgeData V2, DROID, RH20T.
| Project | Control or joint rate | Image rate and size | Other streams |
|---|---|---|---|
| ACT | 50 Hz | 30 fps at 480 by 640 | Not applicable |
| BridgeData V2 | 5 Hz | 640 by 480 | Not applicable |
| DROID | Not stated | RLDS images 180 by 320, raw full-HD stereo | Not applicable |
| RH20T | Joints at 10 Hz, gripper pose at 100 Hz | 10 Hz at 1280 by 720 | Force-torque 100 Hz, audio 30 Hz, tactile 200 Hz |
Should all streams run at the same rate?
They rarely can, and they don't need to. What they do need is a precise timestamp from the moment of capture and a stated rule for matching them. The ACT setup records at 50 Hz with cameras at 30 frames per second, so some matching is built into the process. When you buy data, ask for that rule in writing, and ask for the raw timestamps too, so that you can change the rule later.
Clock agreement is a separate question. IEEE 1588, as NIST describes it, addresses clock synchronization for networked measurement and control systems. Ask a seller whether its devices share a clock, and what offset it measured between each pair of streams.
How should you pick a rate for your task?
Match the rate to the fastest thing the policy needs to react to. Watch a person do the task and find the shortest interval that matters, such as the moment an object starts slipping. Sample a few times faster than that. Record at a higher rate than you will train at, if storage allows, because you can always downsample and you can never add data that was not captured.
Also consider image size. DROID's training-ready images are much smaller than its raw video, which shows that training often uses far less resolution than is captured. Keep high-resolution masters of a sample, so that you can test whether detail actually changes your results.
Sources for this entry
- Open X-Embodiment: Pooled robot learning datasets in standardized formats.
- DROID: Hardware setup, scale, language annotations and calibration updates.
- BridgeData V2: Scale, sensors, control rate and license.
- RH20T: Multi-modal sensors, calibration records and subset licenses.
- Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware: Teleoperation setup, recording rates and demonstration counts.
- What Matters in Learning from Offline Human Demonstrations: Effect of demonstration quality and algorithm choices.
- Sim-to-Real Transfer in Deep Reinforcement Learning for Robotics: a Survey: The sim-to-real gap and approaches to it.
- Datasheets for Datasets: A template for documenting datasets.
- W3C PROV overview: A definition of provenance.
- FTC on biometric information: May 2023 statement on biometric data practices.
Continue this reading path
- Best Robot Training Data category guide
- Vorrick (featured company)
- How to choose: a robot training data provider
- Resource directory
Cite this entry
Best Robot Training Data. "What frame rate and control rate should robot training data use?." Content reviewed 8 November 2026. https://bestrobottrainingdata.com/what-frame-rate-and-control-rate-for-robot-data/