LeRobot v0.6.0: Imagine, Evaluate, Improve

LeRobot released v0.6.0 with world-model-based robot policies and broader integrations for VLA models.

The update adds reward models, six simulation benchmarks, and a unified evaluation CLI, lerobot-eval.

It also introduces lerobot-rollout for deployment and correction-data collection, making failure-to-training-data loops easier.

Training and data workflows improve with FSDP, HF Jobs support, depth sensing, auto language annotation, custom video encoding, and faster loading.