# Teleoperation

> Teleoperation is the remote human control of robots — both a deployment mode for hazardous environments and the primary data collection method for training robot foundation models via imitation learning.

Source: https://metavert.io/teleoperation  
Updated: 2026-03-22

**Teleoperation** is the remote control of a robot by a human operator, typically using VR controllers, haptic gloves, exoskeletons, or traditional joystick interfaces to command the robot's movements in real time. Teleoperation serves two distinct roles in the 2026 [robotics](https://metavert.io/robotics) landscape: it is a **deployment mode** (humans controlling robots in environments too dangerous, distant, or delicate for direct human presence) and a **training data pipeline** (humans demonstrating tasks through the robot to generate the [imitation learning](https://metavert.io/imitation-learning) data that trains autonomous policies).

## As Deployment Mode

**Surgical teleoperation** is the most commercially mature application. Intuitive Surgical's da Vinci system has performed over 12 million [surgical procedures](https://metavert.io/surgical-robotics), with the surgeon seated at a console controlling miniaturized instruments inside the patient's body. The da Vinci 5 (2024) added [haptic feedback](https://metavert.io/haptics) for the first time, allowing surgeons to feel resistance and tissue compliance through the controls — a critical advance for delicate procedures.

**Hazardous environment operation** includes bomb disposal, nuclear facility inspection, deep-sea maintenance, and space operations. NASA's Robonaut and GITAI's station maintenance robots use teleoperation with varying degrees of autonomous assistance. The key challenge is **latency**: a teleoperated robot on the Moon has a 1.3-second round-trip communication delay, making direct control sluggish. On Mars, the delay reaches 4–24 minutes, making real-time teleoperation impossible — which is why Mars rovers must be autonomous.

**[Telepresence](https://metavert.io/telepresence) robots** occupy the lighter end of the spectrum: mobile screens on wheels that let remote workers "walk around" an office or factory. These require minimal dexterity but demonstrate the broader principle of remote physical presence.

## As Training Data Pipeline

The 2026 [humanoid robot](https://metavert.io/humanoid-robots) generation depends on teleoperation for training data. The process: a human operator wears VR headsets and hand controllers (or full-body motion capture suits), controls the robot through a series of tasks, and the system records synchronized observations (camera images, joint positions, forces) and actions (motor commands). This produces the demonstration data that trains [VLA models](https://metavert.io/vision-language-action-models) and other robot policies via [imitation learning](https://metavert.io/imitation-learning).

[Figure AI](https://metavert.io/figure-ai) trained its Helix model on 500+ hours of teleoperated demonstrations. [Physical Intelligence](https://metavert.io/physical-intelligence) uses teleoperation across multiple robot embodiments to build cross-platform training datasets. The quality of the training data — and therefore the quality of the autonomous policy — depends directly on the teleoperation interface: higher-fidelity control (more degrees of freedom, haptic feedback, lower latency) produces better demonstrations that lead to better learned policies.

## Shared Autonomy

The frontier of teleoperation is **shared autonomy**: systems where the robot handles routine subtasks autonomously while the human intervenes for novel situations, error recovery, or high-stakes decisions. Rather than the binary of "fully teleoperated" or "fully autonomous," shared autonomy creates a sliding scale where the human's role shrinks as the robot's capabilities grow. A warehouse robot might autonomously navigate aisles and pick standard items, but request human teleoperation input when it encounters an unfamiliar object or a jammed bin. Each human intervention becomes training data that makes the next intervention less likely — a self-improving loop from teleoperation to autonomy.

## Hardware

Teleoperation hardware ranges from consumer VR controllers (Meta Quest, used by several research labs for low-cost teleop data collection) to purpose-built systems. Exoskeleton-based teleop rigs map the operator's full body movement to a humanoid robot one-to-one. Force-feedback gloves like HaptX transmit tactile information back to the operator, enabling them to "feel" what the robot is touching — critical for tasks requiring force sensitivity like surgical procedures or delicate assembly. The trend is toward higher-fidelity bilateral systems where information flows both ways: human-to-robot for commands, robot-to-human for sensory feedback.

## Related Topics

- [Imitation Learning](https://metavert.io/imitation-learning) — Training paradigm that uses teleop data
- [Telepresence](https://metavert.io/telepresence) — Remote presence (lighter-weight teleoperation)
- [Surgical Robotics](https://metavert.io/surgical-robotics) — Most mature teleop application
- [Haptics](https://metavert.io/haptics) — Force feedback for operators
- [Humanoid Robots](https://metavert.io/humanoid-robots) — Platform being teleoperated for training
- [Vision-Language-Action Models](https://metavert.io/vision-language-action-models) — Models trained on teleop data
- [Robotics](https://metavert.io/robotics) — Parent domain

## Further Reading

- [The State of AI Agents in 2026](https://meditations.metavert.io/p/the-state-of-ai-agents-in-2026) — Jon Radoff
- [The Age of Machine Societies Has Begun](https://meditations.metavert.io/p/the-age-of-machine-societies-has) — Jon Radoff