The useful question is what changes for users, developers or buyers, and whether the announcement stays industry context or becomes something people can actually use.

LeRobot v0.6.0 introduces world model policies such as VLA-JEPA, FastWAM, and LingBot-VA, which learn to predict future outcomes during training but operate without additional inference costs. These policies aim to enhance decision-making by imagining scenarios before acting, addressing a key question in robotics. The release also includes three new world model policies and six additional vision-language-action models (VLAs) like GR00T N1.7 and MolmoAct2, expanding the model zoo for robotics applications.

The update introduces a unified reward models API with Robometer and TOPReward, enabling success detection and progress estimation from raw video and language instructions without task-specific training. Robometer, built on Qwen3-VL-4B, scores task progress using over one million robot trajectories, while TOPReward uses an off-the-shelf VLM to assess trajectories zero-shot. Both models provide labeling scripts for dataset quality inspection and reward-aware behavior cloning.

LeRobot v0.6.0 enhances dataset handling with depth support, automatic language annotations, and custom video encoding options, including hardware-accelerated codecs like NVENC and VAAPI. Depth maps are recorded end-to-end and stored alongside RGB data, while rich language annotations are generated automatically using a VLM. Training efficiency is improved with up to 2x faster data loading, parallel frame decoding, and deterministic sampling for interrupted training sessions.

The release also introduces the lerobot-rollout CLI for deployment with DAgger-style human-in-the-loop corrections, FSDP training for larger models, and cloud training support via Hugging Face Jobs. The codebase has been streamlined, and the installation process is now leaner. The update reflects a focus on closing the robot learning loop by integrating imagination, evaluation, and improvement mechanisms.