LeRobot v0.6.0 Released with New World Model Policies, VLAs, and Reward API
Verified
LeRobot v0.6.0 introduces world model policies (VLA-JEPA, FastWAM, LingBot-VA) and a wave of new VLAs (GR00T N1.7, MolmoAct2, EO-1, EVO1, Multitask DiT). The update adds a reward models API with Robometer and TOPReward, a unified six-benchmark evaluation suite, DAgger-based human correction in the rollout CLI, and FSDP training support. The release requires Python 3.12+, integrates Transformers version 5, and includes support for the Unitree G1 humanoid and NVIDIA IsaacLab-Arena environment.
Sources
- Hugging Face Blog: LeRobot v0.6.0: Imagine, Evaluate, Improve
- Hugging Face Blog: LeRobot v0.5.0: Scaling Every Dimension
- Hacker News: Launch HN: Hebbian Robotics (YC S26) – Build scalable robotics data pipelines
- Hugging Face Blog: `LeRobotDataset:v3.0`: Bringing large-scale datasets to `lerobot`
- Hugging Face Blog: SmolVLA: Efficient Vision-Language-Action Model trained on Lerobot Community Data
- Hugging Face Blog: LeRobot Community Datasets: The “ImageNet” of Robotics — When and How?