What Is LeRobot?
LeRobot is Hugging Face's open-source library for robot learning, the robotics counterpart to what Hugging Face did for language models. It brings pretrained models, shared datasets, and ready-to-run code into one place so anyone can teach a robot to do a task instead of hand-coding every motion.
The core idea is teach-by-demonstration: you show a robot a task by moving it, record that data, and train a model that reproduces the behavior. LeRobot standardizes the messy parts of that pipeline, motor control, data collection, storage, and inference, so a hobbyist and a research lab can run the same tools.
LeRobot is the software anchor of the wider open-source robotics movement covered in our open source robots guide. Where that page explains the movement, this one focuses on the library most projects now build on.
| Piece | What it is | Why it matters |
|---|---|---|
| Models | Pretrained robot-learning policies | Start from a working baseline, not zero |
| Datasets | Shared, standardized demonstration data | Reuse others' data like text or images |
| Code | Motor control, training, inference stack | One pipeline across many robots |
| Hardware | Low-cost arms (e.g. SO-100/SO-101) | Affordable robots the library supports |
How LeRobot Works: Teach By Demonstration
The signature LeRobot workflow uses a leader and a follower arm. A human moves the leader arm through a task; the follower arm mirrors it in real time while cameras record both the motion and the scene. That recording becomes a dataset of what to do and what the world looked like while doing it.
You then train a model on that data and let the follower run the task on its own. Because the data format is standardized, you can combine your recordings with public datasets on the Hugging Face Hub, which is what makes the approach scale beyond one person's demonstrations.
This is the practical, buildable version of the ideas in our embodied AI explainer: intelligence that senses and acts through a body, learned from demonstration rather than scripted. Manipulation quality is still the hard part, as covered in our robotic hands guide.
- Move the leader arm to demonstrate a task.
- The follower arm mirrors it while cameras record the scene.
- Train a model on the recorded demonstrations.
- Run the model so the robot performs the task autonomously.
LeRobot Hardware: The SO-101 And Friends
LeRobot is hardware-agnostic, but it is closely tied to a family of low-cost, 3D-printable arms. The best known is the SO-101 (and its predecessor the SO-100), a 6-DOF open-source arm designed by The Robot Studio with Hugging Face, built to be assembled cheaply and driven directly by LeRobot.
These arms are what make LeRobot tangible for most people: a follower arm costs on the order of a hundred dollars in parts, and a leader-plus-follower pair for teleoperation lands in the low hundreds. That is a fraction of a traditional research arm, which is the whole point.
Hugging Face has also partnered with companies like NVIDIA to add models and simulation tooling to the ecosystem, and there is a free Hugging Face robotics course that walks through setup end to end. The result is a stack where the software, the data, and even the hardware plans are open.
How To Get Started With LeRobot
Getting started is deliberately approachable: install the LeRobot library from its public repository, follow the Hugging Face robotics course, and either buy or 3D print a supported arm such as the SO-101. From there you calibrate the arm, collect a few demonstrations, and train your first policy.
If you have never built a robot arm, start with the hardware first. Our SO-101 robot arm guide covers what to buy and the real cost, and our DIY robot arm kits guide and cost calculator help you budget the full setup.
The honest expectation: LeRobot lowers the barrier a lot, but it does not remove the work. Assembly, calibration, collecting good demonstrations, and training still take real time, and results depend heavily on the quality of your data. It is a genuine on-ramp to robot AI, not a one-click product.
- Install the LeRobot library from its public repository.
- Follow the free Hugging Face robotics course.
- Buy or 3D print a supported arm (SO-100 / SO-101).
- Calibrate, collect demonstrations, then train and run a policy.
Bottom Line
LeRobot is Hugging Face's open-source robot-learning library: pretrained models, shared datasets, and a standardized pipeline that lets anyone teach a robot by demonstration, paired with cheap, printable arms like the SO-101. It is the most accessible on-ramp to real robot AI today, as long as you accept that assembly, good data, and training are still genuine work.
Ready to build? Start with our SO-101 robot arm guide, then use the DIY robot cost calculator to budget a full LeRobot setup.
FAQs
What is LeRobot?
LeRobot is Hugging Face's open-source library for robot learning. It provides pretrained models, shared datasets, and code so anyone can teach a robot tasks by demonstration, and it supports low-cost arms like the SO-101.
Is LeRobot free?
The LeRobot software, models, and datasets are open source and free to use. You still pay for the physical robot hardware, such as an SO-101 arm and cameras.
What hardware does LeRobot use?
LeRobot is hardware-agnostic but is closely associated with low-cost 3D-printable arms like the SO-100 and SO-101, designed by The Robot Studio with Hugging Face, plus cameras for recording demonstrations.
How do you get started with LeRobot?
Install the LeRobot library from its public repository, follow the free Hugging Face robotics course, and buy or 3D print a supported arm such as the SO-101. Then calibrate it, collect demonstrations, and train a policy.
What can you build with LeRobot?
Teach-by-demonstration tasks such as picking, placing, and simple manipulation on a low-cost arm. Results depend on your hardware and the quality of the demonstration data you collect.