{
  "title": "Joint WAM prediction targets",
  "caption": "A joint WAM learns future-world trajectories and action trajectories from observation and task context. The joint training distribution does not specify a universal inference path. Some systems use imagined futures, but others deploy a faster action path shaped by prediction supervision. Author-created explanatory diagram. Playback time is illustrative and is not measured runtime.",
  "alt": "Joint WAM flow: observation and task train world/action branches, without a mandatory world-to-action path during execution.",
  "sources": [
    {
      "path": "content/blog/2026-06-09_world-models-for-embodied-ai/post.md",
      "description": "Article mechanism and retained citations"
    },
    "https://arxiv.org/abs/2605.12090",
    "https://arxiv.org/abs/2603.16666",
    "https://arxiv.org/abs/2607.00836"
  ],
  "limitations": [
    "This playback is an example. Playback duration is not measured runtime.",
    "The two output branches are examples. They do not show the exact head implementation of each architecture.",
    "The diagram adds no future-world-to-action inference dependency or gradient implementation."
  ]
}