Robotics

Teleoperation

ActivePublished: 29 August 2026Updated: 29 August 2026Published

Key innovation

Placing a human in the robot's real-time control loop, enabling remote task execution and the collection of high-quality demonstration data for training manipulation policies.

How it works

The operator moves a master device (mocap glove, exoskeleton, controller, leader arm). Those motions are sensed, mapped into the robot's task or joint space and transmitted over a communication link (often low-latency, e.g. via WebRTC) to a follower robot that reproduces them. The robot's sensors and cameras stream video back and — in bilateral setups — forces and torques, closing the feedback loop. The state-action trajectories recorded during teleoperation form demonstration datasets on which autonomous manipulation policies are trained via behavior cloning / imitation learning.

Problem solved

It lets humans perform dexterous physical tasks where robot autonomy is insufficient or human presence is dangerous, while also solving the data-scarcity problem: it supplies human-driven demonstrations for training manipulation policies that cannot easily be scraped from the internet.

Components

Operator interface (master / leader)Capturing operator motion and commands

Input device capturing the human operator's motions and intent: mocap glove, exoskeleton, controller, leader arm or VR headset.

Communication linkTransporting commands and feedback

Channel transmitting control signals and telemetry between operator and robot with the lowest possible latency (e.g. WebRTC).

Executing robot (slave / follower)Physical task execution

Robot reproducing the operator's motions in the remote environment, equipped with actuators, a gripper/end-effector and sensors.

Feedback (bilateral / telepresence)Closing the perception-action loop

Return of video, forces and torques to the operator, providing a sense of contact with the remote environment.

Official

Implementation

Implementation pitfalls

Latency destabilizes the loopHigh

Communication-link delays degrade bilateral control stability and telepresence quality.

Fix:Low-latency transport (e.g. WebRTC), motion prediction, model-based control.

Operator-robot embodiment mismatchMedium

Kinematic and scale differences between human and robot make faithful motion retargeting hard.

Fix:Motion retargeting, calibration, workspace scaling.

Evolution

Original paper · 1952 · Raymond C. Goertz

Fundamentals of general-purpose remote manipulators

Raymond C. Goertz

1954

Electrical force-reflecting master-slave manipulator

Inflection point

Raymond Goertz at Argonne National Laboratory builds an electrical master-slave manipulator with bilateral force feedback — the birth of modern teleoperation.

2000

Teleoperation in surgery (da Vinci)

Commercialization of teleoperated robotic surgery brings teleoperation beyond the laboratory.

2023

Teleoperation as an imitation-learning data source (ALOHA)

Inflection point

Low-cost bimanual teleoperation rigs (ALOHA) enable collecting demonstrations to learn precise manipulation policies from minutes of data.

2023

Teleoperation data feeds VLA models (RT-2)

Vision-Language-Action models trained partly on teleoperation-collected robot trajectories fuse web knowledge with robot control.