{
  "title": "Dreamer learning and actor deployment",
  "caption": "Dreamer trains the world model from replay, then trains actor and critic on imagined trajectories that start from replay states. Deployment samples the actor after state inference without lookahead search. Author-redrawn mechanism from Hafner et al., without reconstruction images and detailed loss/gradient paths. Source: Hafner et al. Playback time is illustrative and is not measured runtime.",
  "alt": "Dreamer flow: replay trains dynamics and imagined trajectories train actor/critic, then deployment uses live state inference and the actor.",
  "sources": [
    "https://arxiv.org/abs/2301.04104",
    "https://arxiv.org/html/2301.04104v2"
  ],
  "limitations": [
    "This playback is an example. Playback duration is not measured runtime.",
    "The diagram does not include detailed model/behavior loss gradients and optimizer cadence.",
    "Deployment uses recurrent state inference and actor sampling, not lookahead search."
  ]
}