Humanoid Training · Part of The Humanoid Group

Learning From Demonstration

Many humanoid skills begin with a person showing the robot what to do. This guide explains how demonstration and teleoperation data are collected, what makes that data useful, and what it means for your organisation.

By Arjun Rao · Updated October 2026

What learning from demonstration means

Learning from demonstration, often called imitation learning, trains a robot by example rather than by hand-written code. A person performs a task, the scene and the motion are recorded, and a learning model is trained to reproduce that behaviour in new situations. The appeal for humanoids is clear: their bodies are shaped like ours, so human movement translates more directly than it does for other machines. The challenge is generalisation. A policy that copies a demonstration well in one room may struggle when the lighting, objects or layout change, which is why the variety of demonstrations matters as much as the number.

How teleoperation data is collected

Teleoperation puts a person in control of the robot itself, usually through a VR headset, a motion-capture suit, an exoskeleton-style rig or handheld controllers. Because the robot's own joints and cameras record every movement, the data matches the body that will later perform the task. Other approaches record people doing the work directly, with wearable cameras or instrumented gloves, and adapt that footage afterwards. Each method trades cost against fidelity. Teleoperation tends to produce cleaner, robot-specific data but takes operator time; recordings of people are easier to gather in volume but need more processing before a robot can learn from them.

What it means for your organisation

If you plan to deploy humanoids, ask each maker how new tasks are taught and who does the teaching. Some makers collect demonstrations themselves; others expect customers to supply them, or offer tools for your own staff to teleoperate. Ask how many demonstrations a typical task needs, whether data recorded on your site stays yours, and how a learned skill is updated when your process changes. For research teams the questions are practical: which teleoperation hardware a platform supports, how data is logged and exported, and whether the software stack lets you train and run your own policies.

Sources and further reading

Recent Advances in Robot Learning from Demonstration — Annual Review of Control, Robotics, and Autonomous Systems.
Peer-reviewed survey of imitation learning methods, when demonstrating beats scripting, and the problems still open.
Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware — Stanford University and partners (arXiv; RSS 2023).
A low-cost teleoperation set-up plus the ACT imitation method, learning precise two-arm tasks from a small set of demonstrations.
HumanPlus: Humanoid Shadowing and Imitation from Humans — Stanford University (arXiv).
A humanoid copies a person's motion from one RGB camera in real time, and those sessions become data for autonomous skills.

Common questions

Do I need to collect my own demonstration data?

It depends on the maker and the task. Some supply pre-trained skills, while others expect customers to teach site-specific tasks. Ask how teaching works before you shortlist a platform.

What hardware is used for teleoperation?

Common set-ups include VR headsets, motion-capture suits, exoskeleton-style rigs and handheld controllers. Check which ones a platform supports out of the box.

Learning From Demonstration — How humanoids learn skills from human demonstrations and teleoperation data.
Simulation & Sim-to-Real — Why humanoids train in simulation first, and what it takes to cross the gap to the real world.
Evaluation & Readiness — How to judge whether a humanoid is safe, reliable and ready for real work.
Preparing Your Team — The roles, skills and habits your organisation needs before humanoids arrive.
Humanoid Training Services — When to outsource robot data collection, teleoperation and evaluation, and what to ask a provider.
How to Train a Humanoid — A plain-language walkthrough of teaching a humanoid a new task, from demonstration to testing.
Teleoperation for Data — VR, leader-arm and exoskeleton rigs for recording humanoid demonstrations, and what makes a run worth keeping.
Learning to Walk — How reinforcement learning teaches humanoids to walk and balance in simulation before they step out for real.
Training Data and Privacy — Recording people at work or at home to train robots: consent, data protection law and careful data handling.
Choosing First Tasks — How to pick a humanoid's first skills, set pass marks and budget for the demonstrations each one needs.
Safety While Training — Zones, stops and supervision that keep people safe while a humanoid is being taught, tested and debugged.
Latest updates — Teaching robots now takes data operations, simulation and careful testing. These figures and dated updates show where training data comes from, how fast robots learn new tasks and how results are checked.
About us — What we cover, how we source every figure, and who publishes this site.

Visit The Humanoid Group

People also search for robot learning from demonstration, imitation learning robot, programming by demonstration, teaching robots by example, robot teleoperation and teleoperated humanoid robot.