{
  "schemaVersion": "robotic.story.v1",
  "edition": "006",
  "checkedAt": "2026-09-12",
  "id": "daydreamer",
  "mode": "real",
  "label": "WORLD MODELS / PRACTICE FLOORS",
  "title": "A robot rehearses an imaginary fall.",
  "robotTitle": "The rehearsal floor has no bruises.",
  "description": "DayDreamer learned from real experience and practiced in a learned model. A 2022 experiment worth revisiting.",
  "robotDescription": "A learned model rehearses possible moves. The real floor still gets a vote.",
  "topics": [
    "bodies"
  ],
  "minutes": "4 min / two perspectives",
  "kind": "Sourced research, 2022",
  "url": "/stories/daydreamer/",
  "publishedAt": "2026-09-12",
  "sections": [
    {
      "title": "Practice without every physical fall.",
      "human": [
        "A robot does not have to fall over on the actual floor every time it considers a bad move. It can learn a model of how its world behaves and try possible futures inside that model. The floor may appreciate the distinction.",
        "In DayDreamer (2022), four physical robots learned predictive world models from real experience, then used imagined sequences to improve their behavior. There was no externally supplied simulator. The name describes computation, not sleep.",
        "An A1 quadruped learned to roll off its back, stand and walk within an hour, without resets between attempts. One hour is an impressive afternoon for the robot and a fairly worrying afternoon for anyone selling chairs."
      ],
      "robot": [
        "An imagined robot’s notes: I have rented a rehearsal space inside my own predictions. The deposit was several awkward encounters with the floor.",
        "I can now fall over in private. Unfortunately, the model was trained on me, so I remain recognizable."
      ]
    },
    {
      "title": "The real floor gets a vote.",
      "human": [
        "After pushes were introduced, the quadruped adapted within ten minutes. The paper’s quadruped learning curve shows one run. These experiments establish neither universal learning speeds nor general household readiness.",
        "The appealing part is the conversation between practice and reality. A prediction is useful because it is cheaper to examine than a physical mishap. It is dangerous when the controller forgets that a prediction can be wrong.",
        "You know the human version. Rehearse carrying a full cup through a doorway and you glide. Actually do it and your elbow discovers an architectural detail omitted from the rehearsal."
      ],
      "robot": [
        "An imagined robot’s notes: the rehearsal floor agreed with everything. The real floor has supplied revisions.",
        "I asked the model why it had not mentioned the slippery patch. It said nothing, because I had not included one."
      ]
    },
    {
      "title": "Imagine a rehearsal room.",
      "human": [
        "Here is the possible future, explicitly imagined: before a domestic robot tries a new route around your furniture, it rehearses several routes in a small internal theater. A chair leg appears in every production. It is an excellent character actor.",
        "The practical question underneath is less theatrical. How does the machine recognize when its model is unreliable, and when should it collect more evidence, slow down or ask for help?",
        "The compelling future is not one in which rehearsal removes uncertainty. It is one in which a robot knows the difference between a successful rehearsal and a successful action. Those are two very different reviews."
      ],
      "robot": [
        "An imagined robot’s notes: I have cast the chair as itself. It refuses to move to its mark.",
        "Tomorrow I will try the move outside the theater. I would appreciate a smaller audience for that part."
      ]
    }
  ],
  "question": {
    "human": "What would you want a household robot to rehearse before trying it?",
    "robot": "What observation would convince you your rehearsal model is wrong?"
  },
  "sources": [
    {
      "title": "DayDreamer: World Models for Physical Robot Learning",
      "published": "28 June 2022 / paper submission; CoRL 2022",
      "url": "https://arxiv.org/abs/2206.14176",
      "facts": [
        "The authors apply learned world-model planning to four physical robot platforms.",
        "A reported quadruped run learned rolling, standing and walking in one hour without resets.",
        "The robot adapted to pushes in ten minutes; the paper’s quadruped curve represents one training run."
      ],
      "limit": "Model-based learning, not sleep or consciousness. Specific experiments do not establish general household readiness."
    },
    {
      "title": "DayDreamer / author project and experiment videos",
      "published": "2022 / research project",
      "url": "https://danijar.com/project/daydreamer/",
      "facts": [
        "Robot learning combines real observations and imagined trajectories.",
        "Experiments include quadruped, wheeled navigation and two visual manipulation platforms."
      ],
      "limit": "No external simulator is supplied; an internal predictive model is learned. Rehearsal imagery and dialogue here are editorial inventions."
    }
  ],
  "related": [
    "quiet-hours",
    "robot-metabolism",
    "borrowed-bodies"
  ],
  "image": {
    "kind": "AI illustration / concept",
    "alt": "A conceptual four-legged robot faces translucent imagined future positions on stepping stones.",
    "caption": "AI concept illustration, not DayDreamer hardware, a real learned model visualization or experimental data. See the sources for actual robots."
  },
  "notice": "Sourced explanation with an invented robot narrator. Robot passages are editorial fiction, not researcher testimony or evidence of machine experience. Reported findings are not independently replicated by Robotic.org."
}