---
id: lerobot
version: "0.6.1"
license: Apache-2.0
license_treatment: permissive
maintenance: active
---

# lerobot — 🤗 LeRobot: State-of-the-art Machine Learning for Real-World Robotics in Pytorch

License: permissive · Maintenance: active · Downloads: 287.1K/mo

## What it is and what it does

LeRobot is a PyTorch-native robotics framework that standardizes control, data, and training across diverse robot hardware and policy types. It provides a unified Robot class interface that abstracts hardware specifics (supporting platforms from SO-100 arms to humanoids like Unitree G1), a standardized LeRobotDataset format (Parquet + video) hosted on Hugging Face Hub for efficient dataset management, and implementations of state-of-the-art policies spanning imitation learning (ACT, Diffusion, VQ-BeT), reinforcement learning (HIL-SERL, TDMPC), vision-language-action models (Pi0, GR00T, SmolVLA), world models, and reward models.

The framework is designed to lower barriers to entry for robotics research and deployment: you can load pretrained policies, train on standardized datasets with a single command, evaluate on benchmarks like LIBERO and MetaWorld, and extend it with custom hardware or policies by implementing the provided interfaces. It integrates tightly with the Hugging Face ecosystem for model and dataset sharing, making it practical for teams building real-world robotic systems or contributing to shared research infrastructure.

Use it for:

- Train imitation learning policies on collected robot teleoperation data and deploy them to real hardware for manipulation tasks.
- Load and fine-tune pretrained vision-language-action models for new robot platforms without retraining from scratch.
- Standardize data collection and storage across multiple robot instances using the LeRobotDataset format for reproducible research.
- Evaluate robot policies in simulation (LIBERO, MetaWorld) before deploying to physical hardware.
- Implement and share custom robot hardware drivers by extending the Robot interface and leveraging LeRobot's training and evaluation pipeline.
- Benchmark reinforcement learning policies (HIL-SERL, TDMPC) on standardized robotics tasks across different hardware platforms.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

LeRobot provides a unified PyTorch framework for training, deploying, and managing robotic policies and datasets across diverse hardware platforms, from low-cost arms to humanoids.

Yes, if you are building or researching real-world robotics systems. LeRobot is actively maintained, has no known vulnerabilities, and provides a mature ecosystem for policy training and hardware integration. The Apache-2.0 license is permissive for commercial use. Install friction is low and the framework is designed to be extensible. Main consideration: it requires Python 3.12+ and substantial ML dependencies (torch, torchvision); ensure your development environment and target hardware can support them.

## Install

pip install lerobot
uv add lerobot
poetry add lerobot

## Installing lerobot

Before you install: Low friction installation as a pure Python wheel. Active maintenance with recent releases; 16 runtime dependencies including torch, torchvision, and huggingface-hub are substantial but standard for ML robotics work. Requires Python 3.12+.

License in practice: Apache-2.0 permissive license allows commercial and private use with attribution. No copyleft restrictions; you can integrate into proprietary robotics systems.

Quickstart:

pip install lerobot

from lerobot.datasets.lerobot_dataset import LeRobotDataset
dataset = LeRobotDataset("lerobot/aloha_mobile_cabinet")
print(dataset[0]['action'].shape)

Requires Python 3.12 or later; torch and torchvision installation may require system-level dependencies or GPU drivers depending on your hardware target.

Verify before relying:

- Actual GPU/CPU memory requirements for training different policy types (ACT, Diffusion, VLAs, etc.)
- Whether custom robot implementations require C++ extensions or are pure Python
- Real-world transfer success rates for the listed policy types on unseen hardware
- Performance overhead of the standardized Robot interface vs. direct hardware APIs

## Package facts

- License: Apache-2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 287.1K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

robot learning framework pytorch, imitation learning robotics, robot policy training, robotics dataset management, hardware-agnostic robot control, vision-language-action models, reinforcement learning robots, robot simulation and evaluation, robotics, imitation-learning, vision-language-action

[View on SkillFed](https://skillfed.io/packages/lerobot) · [View on PyPI](https://pypi.org/project/lerobot/)