Let’s talk
← Back to home

TELEOP / DATA ACQUISITION

Teleoperation data
on your embodiment.

Human-guided robot operation turns task execution into data. We collect teleoperation datasets through the robot your team needs to study, so observations and actions match the platform you deploy.

Build your dataset How a collection runs

01 / WHAT IT IS

Your embodiment.
Expert guidance.

Teleoperation data is recorded while a human operator controls a robot through a task. The robot’s sensors provide the observations; the operator’s commands provide the actions. Together they form demonstrations in exactly the observation and action space the robot will use when it runs autonomously.

That match is the reason teleoperation datasets remain central to robot learning. Human video and handheld-gripper data teach a model what a task looks like; teleoperation data teaches it what the task looks like from inside the robot, with the robot’s reach, joints, and gripper.

Our teleoperation collections are embodiment-specific. We work with your robot, your control interface, and your task definition, and we scope the operator protocol so that demonstrations are consistent enough to learn from and varied enough to generalise.

02 / WHAT WE CAPTURE

What a teleoperation
collection contains.

Streams depend on the robot and interface. A typical teleoperation collection includes:

Robot observations
Camera streams and proprioceptive state from the robot, synchronised and logged at the rate agreed in the brief.
Operator actions
The commanded actions, in the action space your policy will use: joint targets, end-effector poses, or gripper commands.
Episode structure
Demonstrations segmented to your task definition, with success and failure labelled against the acceptance criteria.
Intervention records
Where the brief covers learning from intervention, the points at which an operator corrected or took over, and what they did.
Variation and site metadata
Object sets, layouts, and environment tags per episode, plus operator identifiers for consistency analysis.

03 / WHAT IT IS FOR

Where teleoperation
data earns its place.

Imitation learning on a specific robot, where demonstrations must be in the robot’s own observation and action space.

Task refinement and recovery: collecting demonstrations of edge cases, regrasps, and corrections that a policy trained on clean data fails on.

Learning from human intervention, using operator takeovers on a partly autonomous system as targeted training signal.

04 / WHERE WE COLLECT

Collected across the
Middle East and Central Asia.

Telepath is based in Dubai and runs collections in real workplaces across the United Arab Emirates and the CIS, with sites in Kazakhstan, Uzbekistan and Tajikistan. The data is built for robotics and physical-AI teams in the United States, led by the San Francisco Bay Area.

Emirates
Dubai, Sharjah and Ras Al Khaimah.
Site types
Laundries, perfume and cosmetics plants, cable factories, furniture factories, plus logistics, hospitality and retail sites.
Languages
Briefs and on-site instructions in English, Arabic and Russian.
Why it matters
Real objects, real layouts and real interruptions, so the variation your model meets in deployment is already in the training data.
Scope a collection Offer a site or record at yours

05 / EXAMPLE TASKS

Briefs we can scope
with this method.

These example briefs suit robot-in-the-loop capture, particularly longer sequences and tasks tied to a specific embodiment. Custom tasks are welcome.

EXAMPLE COLLECTION BRIEFS · CUSTOM TASKS WELCOME

06 / HOW A COLLECTION RUNS

Four stages,
one shared brief.

Teleoperation collections follow our four-stage process, with the Define stage covering robot access, control interface, logging, and operator protocol.

  1. DefineStart with the capability.

    Agree the robot, the control interface, the logged streams and rates, the task definition, the operator protocol, and the acceptance criteria for a successful episode.

    YOUR OUTPUT / A SHARED COLLECTION BRIEF

  2. CollectBring the brief into the world.

    Run sessions with trained operators on the agreed platform and sites. Monitor logging integrity and episode quality during collection, not after.

    YOUR OUTPUT / TASK-ALIGNED DEMONSTRATIONS

  3. ReviewMake every delivery intentional.

    Verify synchronisation, label success and failure, tag interventions and variations, and review coverage against the brief. Schedule re-collection for thin areas.

    YOUR OUTPUT / REVIEWED COLLECTION DATA

  4. DeliverFit your training workflow.

    Deliver observations and actions per episode in your preferred dataset format, with manifest, metadata, and the coverage report.

    YOUR OUTPUT / DATA STRUCTURED FOR YOUR TEAM

FAQ

Teleoperation data, answered.

What is a teleoperation dataset?

A teleoperation dataset is a collection of robot demonstrations recorded while a human operator controls the robot. Each episode pairs the robot’s observations with the operator’s commanded actions, in the same observation and action space the robot uses when running a learned policy.

Do you provide the robot?

Collections are scoped around the embodiment you are training. Robot access, control interface, and logging are agreed at the Define stage. Where you cannot ship a platform to a site, we discuss alternatives during scoping.

How is teleoperation data different from UMI data?

Teleoperation data comes from a real robot under human control and captures the exact embodiment. UMI data comes from a handheld gripper moved by a human hand, with no robot needed. UMI is faster and covers more variation; teleoperation matches the deployment platform precisely.

Can you capture operator interventions on a partly autonomous system?

Yes. Teleoperation data collection can include operator interventions on a partly autonomous system where the brief calls for it. The intervention points and the operator’s corrective actions are recorded and labelled, so they can be used as targeted training signal for the cases the current policy fails on.

What is Telepath’s background in teleoperation?

Telepath’s background in teleoperation comes from previously running remotely operated robot deployments in industrial settings. That operating experience, with real robots, real operators, and real facilities, now goes into scoping and running teleoperation data collections for robot learning teams, including how interventions and recovery behaviour are captured.

FROM THE WORLD. INTO YOUR MODEL.

Teach the robot
from inside the robot.

Build your dataset