Hugging Face Releases LeRobot v0.6.0 Open Robotics Stack

Hugging Face shipped LeRobot v0.6.0, the latest version of its open-source robotics library. The update centers on an 'imagine, evaluate, improve' loop for robot learning — a structured workflow for developing and iterating on robot policies, from simulation/imagination through evaluation to targeted improvement.

LeRobot is Hugging Face's bid to be the open foundation layer for robotics the way its Transformers library became the default for LLMs. Open, shared tooling lowers the barrier for researchers and startups to develop embodied-AI policies without building infrastructure from scratch — a direct extension of Hugging Face CEO Clement Delangue's open-science thesis that shared tooling lets the field 'mutualize' effort and compute.

The timing fits a broader robotics push across the industry: Google DeepMind announced this week that Apptronik's Apollo 2 humanoid platform will collect real-world data to train Gemini Robotics as Apptronik expands its Robot Park facility. Open stacks like LeRobot and proprietary efforts like Gemini Robotics represent two paths to the same embodied-AI goal.

Hugging Face also shipped adjacent updates this week — Transformers v5.13.0 (adding Kimi K2.5-2.7, MiMo-V2-Flash, and a unified exporter) and major updates to its Kernels project for optimized compute. What to watch: adoption of LeRobot v0.6.0 by robotics labs and startups, and whether the imagine-evaluate-improve loop measurably accelerates policy development versus prior versions.