AI
LeRobot v0.6.0: Imagine, Evaluate, Improve
LeRobot v0.6.0, an open-source robot learning library, introduces world-model policies (VLA-JEPA, FastWAM, LingBot-VA) that simulate future states before
Key takeaways
- LeRobot v0.6.0 introduces world model policies (VLA-JEPA, FastWAM, LingBot-VA) that learn to imagine the future before acting
- The release adds a wave of new VLAs (GR00T N1.7, MolmoAct2, EO-1, EVO1, Multitask DiT)
- A new reward models API (Robometer, TOPReward) enables success detection and progress estimation from raw video and language instructions
- Six new simulation benchmarks (LIBERO-plus, RoboTwin 2.0, RoboCasa365, RoboCerebra, RoboMME) are unified under lerobot-eval
- The lerobot-rollout CLI supports DAgger-style human-in-the-loop corrections, turning deployment failures into training data
LeRobot v0.6.0, an open-source robot learning library, introduces world-model policies (VLA-JEPA, FastWAM, LingBot-VA) that simulate future states before acting, a suite of vision-language-action models (GR00T N1.7, MolmoAct2, EO-1, EVO1, Multitask DiT), and a unified reward models API (Robometer, TOPReward) for success detection. The release adds six new simulation benchmarks (LIBERO-plus, RoboTwin 2.0, RoboCasa365, RoboCerebra, RoboMME), depth sensing, VLM-powered dataset annotation, custom video encoding, cloud training via Hugging Face Jobs, FSDP for multi-GPU training, and a deployment CLI (lerobot-rollout) with DAgger-style human-in-the-loop corrections. Installation is streamlined, data loading is up to 2x faster, and the codebase is leaner.
By the numbers
- 0.6.0
- LeRobot version released
- 6
- new simulation benchmarks added
- 2x
- faster data loading speed
- ~5B
- parameters in FastWAM video-generation expert
- 24 - 32GB
- GPU VRAM required for inference
- 10,000
- perturbed variants in LIBERO-plus benchmark
Turn stories like this into views
Ravenclip finds the AI news, makes the video, and posts it before attention moves on.
Common questions
- What happened with LeRobot?
- LeRobot v0.6.0 introduces world model policies (VLA-JEPA, FastWAM, LingBot-VA) that learn to imagine the future before acting
- Where can I read the original report?
- Read the full report at huggingface_blog.
