LeRobot v0.6.0 adds world model policies, new VLAs, reward models API, and dataset improvements

LeRobot v0.6.0 introduces world model policies like VLA-JEPA, FastWAM, and LingBot-VA that learn to imagine the future, alongside a wave of new Vision-Language-Action (VLA) models including GR00T N1.7, MolmoAct2, EO-1, EVO1, and Multitask DiT.

The release also adds a unified reward models API featuring Robometer and TOPReward for success detection, six new simulation benchmarks under lerobot-eval, and dataset enhancements such as depth support, automatic language annotation, and up to 2x faster data loading.

These updates provide researchers with tools to evaluate robot policies more effectively, improve training efficiency through FSDP and cloud support, and access richer, multi-modal datasets for behavior cloning.