Teleops Stations

Where teleoperation sits

Why the top of this ladder is empty

Manipulation data comes from four places, roughly in order of cost: web video, simulation, egocentric capture worn by people, proxy devices like a gripper on a stick — and teleoperation on the robot’s own embodiment, which is the richest form of collection.

Everyone crowds the scalable end. Video and simulation are cheap and infinite, but neither can capture force, and the embodiment gap breaks transfer — a policy trained on a human hand does not run on a two-finger gripper. The high-fidelity corner sits nearly empty only because robots are not deployed at scale yet. Every robot that ships creates demand for tailored, contact-rich data — and that data can only come from a station like these.

The fidelity map

Where the data comes from

Everyone crowds the scalable end. Video and simulation cannot capture force, and the embodiment gap breaks transfer.

Other data sources Station 1 · Base Station 2 · Humanoid Station 3 · Force Lab

Positions are schematic, not measured.

The top rung is not experimental

Station 3 mirrors the setup Google DeepMind (London) and RAI Institute (Boston) independently run for contact-rich manipulation research. It is an established configuration, not a bet.

Configurations

The three stations

Station 1

Base

VR + arms

Data out

Task episodes — video, hand pose, arm joint trajectories, gripper state, time-synced and QA’d. Kinematics only, no force signal.

Good for

High-volume manipulation demos: pick-and-place, sorting, packing. Testing and comparing setups before you commit.

Typical kit

VR headset with tracking
Two position-controlled arms
Dexterous hands or grippers
Clutch-pedal teleop stack
Swappable task cells — shelf, tabletop, bin

Configure this station

Station 2

Humanoid

Mocap suit + full humanoid

Data out

Full-body kinematics — gait, reach and manipulation in one stream, retargeted to your target humanoid.

Good for

Humanoid programmes: whole-body tasks, mobile manipulation, and human-motion pretraining matched to a specific robot.

Typical kit

Xsens Link full-body suit
Humanoid platform, or bring your own
SenseGlove R1 or MANUS gloves
Dexterous hands
Staged environments + retargeting stack

Configure this station

Station 3

Force Lab

Torque-controlled leader-follower

Data out

Torque profiles, contact force, compliance and per-finger pressure alongside full video and kinematics. Force-true demonstrations.

Good for

Contact-rich, fragile and deformable tasks: gripping without crushing, cloth handling, insertion and assembly.

Typical kit

Torque-controlled leader-follower pair
SenseGlove R1 haptics
TESOLLO DG-5F with fingertip force sensors
Licensed mapper + integration pipeline
Contact-rich task cells

Configure this station

Side by side

What each tier gives you

	1 · Base	2 · Humanoid	3 · Force Lab
Kinematics	✓	✓	✓
Video, time-synced	✓	✓	✓
Full-body gait + reach	—	✓	partial
Retargeted to a specific humanoid	—	✓	—
Torque & contact force	—	—	✓
Compliance / per-finger pressure	—	—	✓
Episodes per operator-day	Highest	Medium	Lowest
Relative cost per episode	Lowest	Medium	Highest
Government / defense eligible	Configuration-dependent	Configuration-dependent	Yes, by construction

A note on eligibility

Several strong components in Stations 1 and 2 are China-origin and are offered for commercial use only. If your programme serves a defense or government end customer, tell us at the consult stage — we will configure around it. Station 3 is built entirely from non-China vendors, which is why it is the default for those programmes.

Not sure which rung you need?

Describe the task and the robot. We will tell you the cheapest station that produces the signal your policy actually needs — including if that is the bottom one.

Tell us what you need

Leader devices

Leader arms and teleop controls

The operator side of a station. Leader arms, bimanual kits and clutch controls that drive the follower and record the demonstration.

Why the top of this ladder is empty

Where the data comes from

The three stations

What each tier gives you

Not sure which rung you need?

Leader arms and teleop controls

Trossen WidowX Leader Arm

Clutch Pedal Teleop Control

Franka Torque-Controlled Leader-Follower Pair

Haption Force-Feedback Leader Device

SO-101 Leader & Follower Arm Pair

Trossen AI Bimanual Teleoperation Kit

Compute, sync and tactile sensing

NVIDIA Jetson AGX Thor Developer Kit

NVIDIA Jetson AGX Orin Developer Kit (64GB)

NVIDIA DGX Spark

NVIDIA Jetson Orin Nano Super Developer Kit

Robotiq FT 300-S Force Torque Sensor

Orbbec Multi-Camera Sync Hub Pro

Orbbec Multi-Camera Sync Hub Dev — Star

Orbbec Multi-Camera Sync Hub Dev — Daisy Chain

Orbbec Sync Adaptor — 8-Pin to RJ45

Luxonis FSYNC Y-Adapter

XELA Robotics uSkin uSPa 44 Tactile Patch

XELA Robotics uSkin uSPa 46 Tactile Patch

Tell us what you are building

The rest of the robotics stack