2–3×
Hardware-Synchronized Data Saves You 2–3× Collection Cost
When camera frames, robot state, and actions share a verified clock, fewer demonstrations are discarded and policies learn the intended causal relationship instead of timing noise.
Data Services
Custom Robot Data Collection — Teleoperation, Annotation, and QA at Scale
Expert operators. Professional teleoperation hardware. Your dataset format of choice. From 20-episode pilots to 10,000+ episode production campaigns — we collect the training data your robot policies need.
Leader-Follower TeleoperationVR & Glove CollectionHDF5 / RLDS / LeRobot DeliveryResearch-Grade QA
- Structured packet
- A repeatable episode schema with observations, actions, timing, and metadata.
- State streams
- Robot joints, poses, gripper state, force, and other synchronized signals.
- Real-sim correlation
- Calibration and metadata that make physical data useful for simulation and evaluation.
Why Robot Training Data Quality Matters
Policy performance is bounded by the consistency, timing, and coverage of the demonstrations used to train it. More episodes cannot rescue broken synchronization or inconsistent task execution.
Hard to collect at scale
Robot time is expensive. Reliable throughput requires trained operators, disciplined resets, monitored hardware, and a production schedule.
Operator quality varies wildly
Hesitation, inconsistent motion, and silent task shortcuts create policy artifacts unless operator performance is measured and coached.
Format inconsistency wastes months
Schema drift, missing calibration, and undocumented field semantics turn collection into a long preprocessing project.
How a Data Campaign Actually Works
A managed campaign is an engineering program, not a room full of people pressing record.
- 01
Kickoff Call & Task Design
We define the task, success criteria, scene variation, robot embodiment, sensor stack, and target policy. The result is a collection brief with measurable acceptance criteria.
- 02
Hardware Configuration & Timing Verification
We configure the robot, operator interface, cameras, lighting, and safety limits. Every sensor joins a shared clock before production begins.
- 03
Operator Training & Qualification
Operators rehearse the exact task and pass a proficiency test for success rate, motion quality, consistency, and reset discipline.
- 04
Collection Sprints
Qualified operators collect in controlled sprints while supervisors monitor throughput, failure modes, scene coverage, and per-operator quality.
- 05
Quality Control & Validation
Every episode is checked for task completion, timestamp integrity, frame loss, schema consistency, motion anomalies, and metadata completeness.
- 06
Delivery & Handoff
Approved episodes are packaged with manifests, calibration files, quality reports, and an engineering handoff for immediate training use.
Every Dataset Ships with a Temporal Quality Certificate
A compact, auditable record of whether the delivered streams are safe to train on.
RCSV Temporal Quality CertificateVerified
<5 msCamera-action sync
<1 msCross-camera skew
<0.1%Frame drop rate
<2 msJitter (σ)
<0.5 ms/hrClock drift
Metrics are calculated from delivered episodes, not nominal device specifications. Threshold exceptions are listed explicitly in the handoff report.
Data Specifications
A multimodal episode can include any combination of the streams below, aligned to a common task timeline.
| Signal | Contents | Typical rate | Delivery |
|---|---|---|---|
| Joint states | Position, velocity, effort | 20–100 Hz | Float32 arrays + timestamps |
| RGB cameras | Head, wrist, side, overhead | 30–60 FPS | MP4/JPEG or embedded arrays |
| Depth cameras | Aligned depth + intrinsics | 15–30 FPS | 16-bit depth / point cloud |
| End-effector pose | XYZ + quaternion / 6D pose | 20–100 Hz | Base- and task-frame transforms |
| Gripper state | Position, force, open/close | 20–100 Hz | Continuous or discrete |
| Force/torque | 6-axis wrench and contact events | 100–1,000 Hz | Calibrated SI units |
| Annotations | Success, phases, language, keyframes | Episode / frame | JSON, CSV, or embedded |
All streams carry monotonic timestamps and documented clock provenance. Exact rates and tolerances are fixed in the collection brief.
Data Collection Methods
We choose the operator interface around task precision, throughput, workspace, and embodiment — not around a single favored tool.
| Method | Best for | Precision | Demos/day | Setup | Notes |
|---|---|---|---|---|---|
| Leader-Follower Arms | Precision bimanual and contact-rich tasks | Highest | 30–80 | Medium | ALOHA-style manipulation |
| VR / Quest 3 | Large workspace and mobile manipulation | High | 40–100 | Medium | Natural 6-DOF motion |
| SpaceMouse / Keyboard | Simple Cartesian tasks and pilots | Medium | 60–150 | Low | Fast task validation |
| Haptic Gloves | Dexterous hands and finger control | High | 20–60 | High | Hand pose + tactile tasks |
| Kinesthetic Teaching | Backdrivable arms and precise paths | Highest | 20–50 | Low | Direct physical guidance |
| Scripted Demos | Repeatable baselines and calibration | Deterministic | 100+ | Medium | Coverage and regression checks |
Not sure which interface fits your task? Contact Us and we will scope a pilot.
Operator Quality Assurance Process
Qualification Testing
Every operator passes task-specific trials before production episodes count toward delivery.
Real-Time Monitoring
Supervisors watch success rate, cycle time, frame loss, and scene coverage during collection sprints.
Per-Operator Metrics
Episode quality is attributed by operator so drift and retraining needs are caught early.
Output Formats
Delivered ready for the stack you already use, with schema validation before handoff.
HDF5
Hierarchical episodes with synchronized arrays and video references.
RLDS / TFRecord
TensorFlow-native trajectories with standardized steps and metadata.
LeRobot / Parquet
Hub-ready datasets with Parquet state/action tables and encoded video.
Custom Formats
Your schema, field names, chunking, compression, and validation rules.
Compare tradeoffs in our HDF5 vs RLDS vs LeRobot format comparison guide.
How Many Demonstrations Do You Need?
The answer depends more on task diversity and policy objective than on a universal episode count.
| Task class | Example | Typical demos | Goal | Variation |
|---|---|---|---|---|
| Simple single-arm | Pick-and-place, one object | 20–50 | Pilot / behavior cloning | Low |
| Moderate single-arm | Variable objects and poses | 100–300 | Robust task policy | Medium |
| Bimanual | Coordinated assembly or handoff | 300–1,000 | ACT / diffusion policy | High |
| High diversity | Many objects, scenes, operators | 1,000–5,000 | Generalization | High |
| VLA / generalist | Language-conditioned multi-task | 5,000+ | Foundation-model adaptation | Very high |
Pricing
Start with enough data to validate the task and instrumentation, then scale only after the pilot passes.
20 demonstrations
$2,500- Task design and collection brief
- Hardware setup and timing validation
- Expert operator collection
- QA report + one delivery format
- 1–2 week turnaround
- 2 weeks post-delivery support
Campaign · 100 demonstrations
$8,000- Everything in Pilot
- Multi-station parallel collection
- 2–4 dedicated operators
- Weekly batch deliveries
- Scene diversity management
- Up to 2 delivery formats
Custom scale / ongoing
Custom- Dedicated collection infrastructure
- On-site or co-located deployment
- SLA with uptime guarantees
- Custom robot integration
- All formats + Platform access
- Ongoing support and iteration
Compatible Hardware
If your platform is ROS2-compatible, we can usually collect data on it. Custom integrations are scoped before production.
See our full hardware catalog for specifications and availability. Leasing rates are available for supported platforms.
10-Point Data Quality Checklist
Every episode we deliver passes this checklist. No exceptions.
- 1
Synchronized timestamps — All cameras, joints, and actions aligned to <5 ms tolerance using shared clock sources.
- 2
Consistent episode structure — Identical observation/action schema, dimensions, data types, and key names.
- 3
Operator qualification — Operators pass a proficiency test on the specific production task.
- 4
Task success verification — Every episode is reviewed; failures are flagged and separated on request.
- 5
Scene reset consistency — Object positions, lighting, and workspace state follow documented reset ranges.
- 6
Frame drop monitoring — Camera streams are checked and episodes above the allowed loss rate are recollected.
- 7
Gripper state consistency — Gripper signals are validated against visible contact and commanded actions.
- 8
Joint limit compliance — Trajectories are checked for limit violations, clipping, and unsafe motion.
- 9
Metadata completeness — Robot, operator, task, calibration, scene, and timing metadata ship with every episode.
- 10
Annotation standards — Labels follow a documented ontology with spot checks for consistency.
Campaign Examples
2,400 bimanual demos in 6 weeks
Two qualified operator teams collected coordinated manipulation episodes with weekly HDF5 deliveries and timing reports.
First policy in 4 weeks
A 100-demo pilot validated the task, then expanded into a production dataset packaged for LeRobot training.
Ongoing pipeline, 72% to 94% success
Failure mining informed recurring collection sprints and targeted variation for each policy release.
5,000-episode benchmark dataset
A reproducible multi-task corpus with held-out scenes, annotations, manifests, and public benchmark splits.
Who Uses RCSV Data Services
Research Labs
Publish reproducible results with documented procedures, timing health, and benchmark-ready splits.
Startup Policy Training
Get from task definition to a trainable pilot without building an internal operator program first.
Enterprise Deployment
Scale ongoing collection with governance, SLAs, private handling, and repeatable QA.
Academic Benchmarks
Create multi-institution datasets with shared schemas, calibrated hardware, and consistent annotations.
Trusted by Leading Research Institutions
StanfordUC BerkeleyMITCMUToyota Research
Frequently Asked Questions
What teleoperation hardware do you use?
We operate leader-follower arms (ALOHA-style WidowX/ViperX and OpenArm setups), Meta Quest 3 VR systems, 6-DOF SpaceMouse interfaces, and SenseGlove Nova 2 haptic gloves. We select the interface that best matches your task requirements for precision, throughput, and data quality. For bimanual tasks, we run dual leader-follower or dual VR configurations.
What formats do you deliver?
We deliver datasets in HDF5 (ACT/ALOHA compatible), RLDS/TFRecord (for Open X-Embodiment and Octo), LeRobot Parquet (Hugging Face Hub ready), or custom formats. You specify the format in your project brief, and we handle conversion and validation.
How long does a data collection campaign take?
A pilot program (20 demos) typically takes 1–2 weeks from kickoff to delivery, including task design and hardware setup. A standard campaign (100 demos) takes 2–6 weeks depending on task complexity and scene diversity requirements. Enterprise-scale projects are scoped individually.
Can you collect data on my robot?
Yes. We work with OpenArm, DK1, Franka FR3, UR3e, UR5e, xArm, Kinova Gen3, Unitree G1, and most ROS2-compatible robot arms. If you ship us your robot or we can procure one, we integrate it into our collection infrastructure. Custom integrations typically take 3–5 business days.
What is a typical cost per episode?
Cost per episode ranges from $8–$35 depending on task complexity, number of camera views, teleoperation method, and QA requirements. Simple tabletop tasks are at the lower end; contact-rich bimanual tasks with dexterous hands are at the higher end. Volume discounts apply for campaigns over 500 episodes.
Do you sign NDAs?
Yes. We sign mutual NDAs before project discussions involving proprietary tasks, robot configurations, or research goals. Data collected under contract is owned by the client. We do not use client data for other purposes or include it in public datasets.
Can collected data go directly into the Fearless Platform?
Yes. Enterprise data collection campaigns can include Fearless Platform access. Collected data can flow into your workspace with metadata, QA reports, and lineage information for replay, annotation, evaluation, and retraining.
What annotation types are available?
We support timestamped task phases, segmented subtask boundaries, language instructions for VLA training, keyframe annotations, and success/failure labels. Custom annotation schemas are supported for enterprise campaigns.
Related Resources
Guide
Robot Data Collection Guide
Teleoperation methods, station setup, operator training, and quality assurance.
Read more GuideHDF5 vs RLDS vs LeRobot Formats
Technical comparison with conversion examples and pipeline compatibility.
Read more GuideBimanual Teleoperation Setup
Hardware configuration and calibration for dual-arm collection.
Read more PlatformFearless Data Platform
Manage data, replay episodes, mine failures, and close the loop to retraining.
Read more HardwareHardware Catalog
OpenArm, DK1, dexterous hands, UR series, and more.
Read more EnterpriseEnterprise Programs
Managed programs, fleet deployment, and SLA-backed operations.
Read more ResearchRobot Time Fabric
Why temporal alignment matters for robot learning data efficiency.
Read moreReady to Start Your Data Collection Campaign?
Tell us the robot, task, sensor stack, and target policy. We will turn it into a concrete pilot scope.
contact@roboticscenter.aiTeleop Dataset Program
Build your dataset scope
Share the basic project context and technical streams you need. The beta site sends the complete scope directly to our data operations team for a tailored quote.
- Full technical handoff in one request
- No commitment required
- Response within one business day