01Teleop: Hand + Glove
A tactile glove drives the robot hand while we record fine-finger teleoperation data.
- Data
- Finger pose, joint angles, tactile signals, and robot state.
- Best for
- Fine manipulation, contact-rich skills, and grasp diversity.
Dexterous Robot Hand Data Collection
Collect teleoperation data, Universal Manipulation Interface (UMI) data, and egocentric demonstrations for dexterous robot hands—single-hand or bimanual.
Single-hand & bimanual Your hand or ours San Francisco lab
Collection methods
Teleoperation, egocentric, and UMI data collection methods
Compare glove-based hand teleoperation, VR arm-and-hand teleoperation, egocentric capture, and UMI workflows for your task.
0102
Teleop: Arm + Hand, Glove + VR
VR tracks the arm while a glove controls the fingers, producing synchronized arm-and-hand teleoperation data.
- Data
- Arm pose, hand joints, actions, video, tactile, and robot state.
- Best for
- Whole-task demonstrations with reaching, grasping, and placement.
03Egocentric: Head Camera + Tactile Glove
A person performs the task while first-person video and touch are recorded.
- Data
- Egocentric video, hand pose, tactile signals, and task labels.
- Best for
- Learning human strategies and contact cues from natural motion.
04
Egocentric: Head Camera Only
A lightweight first-person setup captures human task demonstrations at volume.
- Data
- Egocentric video, timestamps, session metadata, and task labels.
- Best for
- Lower-cost, high-volume human video and broad task coverage.
05
UMI (Universal Manipulation Interface) data collection
A tracked handheld gripper records portable demonstrations for transfer to a robot setup.
- Data
- Camera video, gripper state, tracked pose, and task labels.
- Best for
- Fast, scalable manipulation data before a full robot rig is ready.
Need the hardware too?
Buy the same hands and gloves used for collection.
Keep your training and deployment embodiment aligned with hardware sourced through RCSV.
Qualified scope
Tell us what you want to collect.
Share the method, hardware, configuration, and rough hours. We’ll reply with the right pilot scope and matching paperwork.
Prefer email?contact@roboticscenter.ai
Data in action
Real demonstrations, captured in motion
See representative egocentric workflows across food preparation, object organization, and garment handling.
What you get
Learning-ready data with a documented handoff
Every project starts with an agreed stream list and ends with a format and quality report your team can use immediately.
Synced multi-view video
Head, wrist, side, or overhead views aligned to the task timeline.
Hand pose and joint angles
Finger configuration and embodiment state for every demonstration.
Tactile signals
Contact and touch streams when the selected glove or hand supports them.
Robot state
Arm, hand, gripper, pose, and action streams defined in the project brief.
Task labels
Success, failure, phase, and custom labels agreed during scoping.
How it works
Pilot the setup. Then scale with confidence.
A four-step path keeps technical risk at the front and gives your team a concrete checkpoint before production collection.
- 01
Scope call
Choose the task, method, hardware, views, signals, and acceptance criteria.
- 02
Pilot batch
Validate operator workflow, synchronization, schema, and task quality on a small batch.
- 03
Full collection
Scale the approved setup with trained demonstrators and monitored sessions.
- 04
Quality check & delivery
Review the data, package the selected format, and hand off the QA report.
FAQ
Questions before a scope call?
Quick answers about scope, timing, formats, and visiting the lab.
Start a quoteWhat is dexterous hand teleoperation data?
Dexterous hand teleoperation data records a human operator controlling a multi-finger robot hand. A dataset can include synchronized video, finger joint states, actions, tactile signals, and robot state for imitation learning.
What is UMI data collection?
Universal Manipulation Interface (UMI) data collection uses a tracked handheld gripper and camera to capture portable manipulation demonstrations, including video, gripper state, pose, timestamps, and task labels.
What is the minimum order?
Minimum size depends on the collection method, hardware, and task complexity. We recommend a pilot large enough to validate the workflow before full collection.
What is the typical lead time?
Most pilots take 1–2 weeks. Full collection programs typically take 2–6 weeks; complex bimanual setups may take longer.
Can you collect a custom task?
Yes. We define the task steps, variations, reset conditions, success criteria, and labels, then validate the workflow in a pilot.
Which data formats can you deliver?
We deliver HDF5, RLDS/TFRecord, LeRobot/Parquet, or an agreed custom schema. Every delivery includes validated structure and metadata.
Can our team visit the lab?
Yes. Schedule a visit to our data collection lab at 90 Welsh St in San Francisco to review hardware or scope a pilot.