https://k4i.top/tags/jepa/Recent content in Jepa on k4i's blogHugo -- gohugo.ioensky_io@outlook.com (K4i)sky_io@outlook.com (K4i)All content is subject to the license of <a rel="license noopener" href="https://creativecommons.org/licenses/by-nc-sa/4.0/" target="_blank">CC BY-NC-SA 4.0</a> .Thu, 18 Jun 2026 10:00:00 +0800https://k4i.top/posts/embodied-models-vla-jepa-wam/Thu, 18 Jun 2026 10:00:00 +0800sky_io@outlook.com (K4i)Thu, 18 Jun 2026 10:00:00 +0800https://k4i.top/posts/embodied-models-vla-jepa-wam/<p>If a language model only has to answer with text, an embodied model has to answer one extra question: <strong>what should this sentence become as an action?</strong></p> <p>Suppose you tell a tabletop robot: “push the red cup next to the plate.” The model must identify the cup, understand “next to,” decide how the arm should move, close or release the gripper at the right moment, and recover if the cup slips. The hard part is not multimodality alone. It is the closed loop between language, vision, physical state, and continuous action: the action changes the world, and the new world changes the next action.</p>K4ifeatured imageembodied-airoboticsvlaworld-modeljepanotes