00//Teleoperation / Robot Learning

Teleop Pro

Teach Robot Arms by Demonstration.

One platform for the whole robot-learning workflow: hardware integration, data collection in VR, safety, dataset cleaning and sync, training, inference, and remote assistance when a policy fails.

Teleop Pro / Demonstration
Intro//What It Is

From a Demonstration to a Policy.

Teleop Pro is the software between your robot arm and a robot-learning model. People show the arm a task, Teleop Pro turns that into training data, and the trained policy runs on the same arm, with a person able to take over at any moment.

VR teleoperation / Real time

One Platform for the Whole Workflow.

Teaching a robot arm a task takes much more than a model. Teleop Pro covers all eight stages, from connecting the hardware to helping the policy when it fails.

01//Hardware Integration

Connected to Your Arm and Cameras.

Drivers for the arm, its cameras and the headset, and the services that tie them together on a PC beside the robot. No ROS 2 required.

Works today with
  • AgileX Piper
  • UFACTORY Lite6
  • Intel RealSense
  • ZED Mini
  • Meta Quest
  • Pico 4
Another arm or camera? We integrate it with you.
  • One PC beside the robotRuns as a set of services, without ROS 2
  • In the browserThe VR app and the TV workspace need no install
  • Your hardwareA new arm needs a driver, a kinematic model and a config
02//Data Collection / Teleoperate

Collect Data by Showing the Task.

An operator drives the arm in VR: squeeze to move, pull the trigger to grasp, let go to rest your hand.

VR controller / Real time

Precise Enough for Cotton Swabs.

Here an operator picks up cotton swabs one at a time. The arm follows the hand at 200–300 Hz with low latency, so small corrections land where you intend.

You See Depth, Not a Flat Screen.

A stereo camera over the workspace sends one image to each eye. You judge how far the gripper is from an object the way you do with your own hands, which is what makes fine tasks easy to teleoperate.

  • In the browserWebXR on the headset, nothing to install
  • Depth perceptionZED stereo camera, one image per eye
  • 200–300 Hz controlInverse kinematics on the robot PC
  • Guided practiceNew operators learn the controls on a virtual gripper
02//Data Collection / Record

Three Ways to Record.

Drive every episode yourself, or let the recorder repeat a motion while you only move the objects.

RecorderPick & place
The recorder's Pick & place screen
Pick & placeGenerated · stops at every point
In every episode
  • RGB video
  • Depth in mm
  • Joint positions
  • TCP pose
  • Joint efforts
  • Gripper
  • Commanded actions
  • Task instruction
03//Safety

Safe to Hand to an Operator.

One safety layer checks every target, whether it comes from a person, the recorder or a policy.

  • No-go zonesTargets inside a commissioned zone are rejected before the arm moves
  • AbortPress and hold: the arm goes home and stays until you release it
  • Hold to moveCommands flow only while you hold the control; let go and the arm stops
  • Operator firstYour commands outrank the recorder and the policy
04//Data Cleaning

Clean the Dataset Before You Train.

A bad demonstration teaches a bad habit. Replay every episode, see exactly what was recorded, fix the instruction, and delete the takes that went wrong.

Teleop ProDataset manager
Teleop Pro's dataset manager with the list of episodes, the task prompt and the playback of a pick-screwdriver episode recorded in our lab: scene camera, gripper camera and depth

One Episode, as It Is Stored.

Episode 3 of 135 / pick-screwdriverLeRobot format
Gripper camera · RGB
Depth · mm

Task“pick screwdriver”

  • ReplayCameras, depth and every signal, in sync
  • Fix the instructionEdit the task prompt of any episode
  • Delete bad takesRemove an episode; the rest are renumbered
  • LeRobot formatWorks with the tools you already train with
05//Sync

Local First, Shared When You Choose.

Episodes are saved on the robot PC. One press pushes a private copy to Hugging Face, and the machine that trains pulls the same dataset.

  • Local firstRecording never waits for an upload
  • PrivatePushed to a private dataset on your Hugging Face account
  • Push and pullBoth are explicit, from the dataset manager
06//Training / SpesVLA Research Preview

Train on Your Data, With Our Model.

SpesVLA is our vision-language-action model: camera images and a sentence in, arm motion out. We train it on your episodes. The dataset is standard LeRobot, so your own training code works too.

95%LIBERO-Long success rate

On a standard long-horizon benchmark SpesVLA scores with the best open models at a fraction of their size.

Read the Research Preview
  • Your data inThe episodes you recorded and cleaned
  • Our modelSpesVLA, tuned to your arm, cameras and task
  • Or your ownStandard LeRobot format: bring your training code
LIBERO-Long success rate (%)Model size

SpesVLA: internal simulation benchmark, final checkpoint of the latest run, 10 trials per task (100 in all); the papers use 50 per task. The best intermediate checkpoint reached 100/100; a 50-trial confirmation is pending. Others as published: OpenVLA-OFT (Kim et al. 2025), π0.5 (openpi), GR00T N1 (as reported by Wang et al. 2025), SmolVLA (Shukor et al. 2025).

07//Inference

The Policy Runs on the Same Arm.

The trained policy drives the arm through the same interfaces and the same safety layer the operator used. It watches the cameras and re-plans as the scene changes.

Someone moves the screwdriver mid-grasp. The policy follows. / Real time
“Pack files and screwdrivers”: one sentence, two trays.
  • Closed loopRe-plans from the camera as things move
  • One sentenceThe instruction picks the task
  • Same safety layerNo-go zones and Abort apply to the policy too
08//Remote Assistance

When It Fails, Take Over. Then It Learns.

No policy is right every time. When it fails, squeeze the grip: the arm is yours mid-motion, from the lab or over the internet. Finish the task, and the fix becomes training data for the next model.

  • Instant takeoverThe operator outranks the policy; no restart, no mode switch
  • From anywhereAssist over the internet, in VR
  • Recovery is dataRecord the fix as an episode and train on it
10//Open Source & Commercial

Teleop or Teleop Pro?

Open source

Teleop

A library: WebXR motion input for your own robot code. You build the rest.

  • Interfaces for xArm and ROS 2
  • LeRobot teleoperator plugin
Teleop on GitHub
Commercial

Teleop Pro

The whole workflow, set up on your hardware by our engineers.

  • All eight stages above, in one platform
  • VR teleoperation with live stereo video
  • Safety layer, recorder and dataset manager
  • SpesVLA research: policies trained on your data
  • Remote takeover when a policy fails
Discuss Your Setup
11//Questions

Planning Your Setup.

What do we need to start?
A supported arm and camera, a PC beside the robot, and a VR headset. The operator works in the headset's browser; there is nothing to install on it.
Can we use our own robot?
Yes. A new arm needs a device driver, a kinematic model and a deployment config; we build them with you.
What happens when the policy fails?
An operator takes over: their commands outrank the policy, so squeezing the grip hands them the arm at once, on site or over the internet. The recovery can be recorded and added to the training data.
Where does our data live?
On your machine, in LeRobot format. Pushing it to Hugging Face is a separate step you take when you choose.
Do we need ROS 2?
No. Teleop Pro runs without ROS 2; the open-source Teleop library has a ROS 2 interface if you want one.

Teleop Pro is partly funded by the Innovation Fund of the Republic of Serbia.

Republic of Serbia Ministry of Science, Technological Development and Innovation, Innovation Fund, and Expo 2027 Belgrade Serbia

Show Us the Task You Want to Teach.

Tell us about your arm and the task; we will show you the whole workflow on it.

Talk to Our Engineers