LeRobot v0.6: open-source robotics framework gains imagination, evaluation, and continuous improvement
LeRobot v0.6: the framework that closes the AI robotics loop
Hugging Face has released LeRobot v0.6.0, likely the most significant update to the open-source AI robotics framework yet. The tagline sums it up: "Imagine, Evaluate, Improve".
What's new
World Models: Three new architectures — VLA-JEPA, FastWAM, and LingBot-VA — that learn to imagine the future before acting. The robot mentally simulates consequences before executing movements.
Reward Models: Robometer and TOPReward automatically evaluate whether an action was successful, without human intervention.
Continuous Improvement: The new lerobot-rollout CLI implements DAgger-style corrections: when the robot fails, the human corrects, and that correction becomes training data immediately.
Why it matters
AI-powered robotics is becoming accessible to developers and researchers without multi-million dollar labs. LeRobot v0.6 democratizes the complete robot learning loop: simulate, train, evaluate, correct, improve — all open-source.
LeRobot v0.6 is available on GitHub and Hugging Face Hub.