LeRobot

license

Apache-2.0(LICENSE body)

source

public

core features withheld

no — no enterprise path in the tree and no paid tier in the README

The whole library is Apache-2.0, and nothing in the repository is set aside for a paid edition. The policies it ships are the library's own code; pretrained weights it can load, such as the pi0 ports, carry their own licenses and are separate entries.

• https://raw.githubusercontent.com/huggingface/lerobot/main/LICENSE recorded 2026-09-26

  LICENSE: "Copyright 2024 The Hugging Face team. All rights reserved." followed by the Apache License, Version 2.0 text; appended notices for third-party code are all MIT or Apache.

• https://raw.githubusercontent.com/huggingface/lerobot/main/README.md recorded 2026-09-26

  README: policies, dataset tooling and hardware support are all in the repository; no hosted service, paid plan or withheld feature is mentioned.

• https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/huggingface%2Flerobot recorded 2026-09-26

  Repository record for huggingface/lerobot: license apache-2.0, archived false, fork false, pushed 2026-09-22.

• https://ungh.cc/repos/huggingface/lerobot/files/main recorded 2026-09-26

  Full file listing of the main branch, 1,150 paths; none sits under an enterprise, ee or commercial directory.

PyPI installs of the lerobot package, which the project's own metadata links to this repository. Installs from a source checkout, the route most hardware builders follow, are not counted.

• https://pypi.org/pypi/lerobot/json recorded 2026-09-26

  PyPI metadata: project_urls source https://github.com/huggingface/lerobot; license Apache-2.0.

• https://pypistats.org/api/packages/lerobot/recent recorded 2026-09-26

  pypistats: last_month 230,678 downloads for lerobot.

The software in this category is compared on how fast and how faithfully it simulates robots. LeRobot runs no physics of its own: it collects data on real hardware, trains policies, evaluates them in other projects' benchmark environments and runs them back on the robot, so it has no place on that comparison and none is claimed.

• https://raw.githubusercontent.com/huggingface/lerobot/main/README.md recorded 2026-09-26

  README policy table: imitation learning (ACT, Diffusion, VQ-BeT, Multitask DiT), reinforcement learning (HIL-SERL, TDMPC), VLAs (Pi0, Pi0Fast, Pi0.5, GR00T N1.7, SmolVLA, XVLA, EO-1, MolmoAct2, WALL-OSS, EVO1); "a hardware-agnostic, Python-native interface ... from low-cost arms (SO-100) to humanoids"; evaluation runs in external benchmark environments such as LIBERO and MetaWorld, and no physics engine of its own is described.