LeRobot v0.6.0: Imagine, Evaluate, Improve

LeRobot v0.6.0 introduces features to close the robot learning loop, including world model policies like VLA-JEPA and LingBot-VA that imagine future outcomes before acting. The release adds new Vision-Language-Action (VLA) models, reward models such as Robometer, and a unified command-line interface for benchmarks. Dataset capabilities are enhanced with depth support, automatic language annotation, and up to 2x faster data loading. Training and deployment are improved through a new rollout CLI, FSDP support for large models, and cloud training integration.