{
  "world-models-2018": {
    "kind": "model",
    "glossary": {
      "output": "潜在向量是模型内部的编码。这里输出的是描述预测分布的参数。"
    },
    "input": {
      "value": "记忆模型 M 接收当前潜在向量和动作。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "用于采样下一潜在向量预测的高斯混合分布参数。",
      "tags": [
        "latent"
      ]
    },
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "games"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:input"
          ]
        },
        "domains": {
          "games": [
            "direct:game-environment"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "sources": {
        "game-environment": {
          "url": "https://worldmodels.github.io/",
          "quote": "Our VAE trained on observations from CarRacing-v0 [47].",
          "observed_on": "2026-09-21",
          "snapshot_hash": "bddb485f1997cf4dc02fde117a7d9a2c880c683b4a7c54ae5d92e214c2c849d2"
        }
      }
    }
  },
  "sora-world-simulators": {
    "kind": "model",
    "glossary": {
      "output": "2024 年报告说明视频时长可变，最长一分钟；这里的一分钟是上限。"
    },
    "input": {
      "value": "图像和提示词。",
      "tags": [
        "visual",
        "text"
      ]
    },
    "output": {
      "value": "一分钟视频。",
      "tags": [
        "visual"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "language"
      ],
      "domains": [
        "unknown"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "technical:input"
          ]
        },
        "domains": {}
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "本页限定为 2024 技术报告。文本条件不表示已确认逐步动作控制；应用领域暂不从展示片段推断。"
    }
  },
  "copilot4d": {
    "kind": "model",
    "glossary": {
      "input": "BEV 是俯视角表示；token 是模型处理的编码单元。",
      "output": "点云是用三维空间中的点表示物体或场景的数据。"
    },
    "input": {
      "value": "BEV token 和自车动作。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "由预测的 BEV token 解码得到的未来点云。",
      "tags": [
        "spatial"
      ]
    },
    "classification": {
      "types": [
        "latent",
        "simulation"
      ],
      "controls": [
        "driving_action"
      ],
      "domains": [
        "driving"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:route"
          ],
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "driving_action": [
            "technical:input"
          ]
        },
        "domains": {
          "driving": [
            "direct:driving-study"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "核心模型预测 BEV token，再解码为未来点云；两种标签分别对应核心预测与传感器模拟结果。",
      "sources": {
        "driving-study": {
          "url": "https://waabi.ai/research/copilot-4d",
          "quote": "Copilot4D: Learning unsupervised world models for autonomous driving via discrete diffusion",
          "observed_on": "2026-09-21",
          "snapshot_hash": "185249c591af3dca9e11b355f5259e8203b2200c5886657f70d79368b8f21698"
        }
      }
    }
  },
  "genie-3": {
    "kind": "model",
    "glossary": {},
    "input": {
      "value": "导航输入和文本交互。",
      "tags": [
        "action",
        "text"
      ]
    },
    "output": {
      "value": "可实时导航的动态视觉世界。",
      "tags": [
        "visual"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "navigation",
        "language"
      ],
      "domains": [
        "unknown"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "navigation": [
            "technical:input"
          ],
          "language": [
            "profile:genie-3-context"
          ]
        },
        "domains": {}
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "导航和文本事件是不同控制方式；生成环境的展示不自动限定为游戏或机器人应用。"
    }
  },
  "vjepa-2-ac": {
    "kind": "variant",
    "glossary": {
      "input": "末端执行器指机械臂末端执行任务的部件，例如夹爪。",
      "output": "“表征”可以理解为模型内部对画面信息的编码，并非直接输出可观看的视频。本体感知指机器人对自身状态的感知。"
    },
    "input": {
      "value": "过去的视频帧、动作和末端执行器状态。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "以控制动作和本体感知观测为条件，预测未来视频观测的表征。",
      "tags": [
        "latent"
      ]
    },
    "scope_note": "本页聚焦 V-JEPA 2-AC，接口与限制不自动代表 V-JEPA 2 基础模型。",
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "robot_action"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "robot_action": [
            "technical:input"
          ]
        },
        "domains": {
          "robotics": [
            "technical:route"
          ]
        }
      },
      "reviewed_on": "2026-10-01"
    }
  },
  "marble": {
    "kind": "model",
    "glossary": {
      "output": "Gaussian splats（高斯泼溅）：用大量半透明粒子表示三维场景的一种方式。"
    },
    "input": {
      "value": "文本、图像、视频或粗略 3D 布局。",
      "tags": [
        "text",
        "visual",
        "spatial"
      ]
    },
    "output": {
      "value": "可导出为 Gaussian splats、网格或视频的 3D 世界。",
      "tags": [
        "spatial",
        "visual"
      ]
    },
    "classification": {
      "types": [
        "spatial"
      ],
      "controls": [
        "language",
        "scene_edit"
      ],
      "domains": [
        "spatial_creation"
      ],
      "basis": {
        "types": {
          "spatial": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "technical:input"
          ],
          "scene_edit": [
            "technical:route"
          ]
        },
        "domains": {
          "spatial_creation": [
            "technical:one_liner"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "生成三维空间与导出视频分别理解；场景编辑不等于已确认动作后果预测。"
    }
  },
  "gwm-1": {
    "kind": "family",
    "glossary": {},
    "input": {
      "value": "家族所述动作控制包括相机位姿、机器人指令和音频。",
      "tags": [
        "action",
        "audio"
      ]
    },
    "output": {
      "value": "逐帧生成的视觉序列。",
      "tags": [
        "visual"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "camera",
        "robot_action",
        "audio"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "camera": [
            "technical:input"
          ],
          "robot_action": [
            "technical:input"
          ],
          "audio": [
            "technical:input"
          ]
        },
        "domains": {
          "robotics": [
            "profile:gwm-1-context"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "控制标签来自家族材料；机器人领域对应 GWM Robotics，不代表 Worlds、Avatars 支持全部接口或用途。"
    }
  },
  "waymo-world-model": {
    "kind": "model",
    "glossary": {},
    "input": {
      "value": "语言提示、驾驶输入和场景布局。",
      "tags": [
        "text",
        "action"
      ]
    },
    "output": {
      "value": "包含相机与激光雷达数据的多传感器模拟输出。",
      "tags": [
        "visual",
        "spatial",
        "sensor"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "language",
        "driving_action",
        "scene_edit"
      ],
      "domains": [
        "driving"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "technical:input"
          ],
          "driving_action": [
            "technical:input"
          ],
          "scene_edit": [
            "technical:input"
          ]
        },
        "domains": {
          "driving": [
            "technical:input"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "包含相机和激光雷达模拟；场景布局条件不自动代表可导出的三维场景生成。"
    }
  },
  "cosmos-3": {
    "kind": "family",
    "glossary": {},
    "input": {
      "value": "语言、图像、视频、音频和动作序列。",
      "tags": [
        "text",
        "visual",
        "audio",
        "action"
      ]
    },
    "output": {
      "value": "语言、图像、视频、音频和动作序列。",
      "tags": [
        "text",
        "visual",
        "audio",
        "action"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "physical_ai"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "physical_ai": [
            "technical:route"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "标签概括家族材料中明确的视觉生成与动作条件；各变体接口分别核对，不将全部模态归为统一控制方式。"
    }
  },
  "gaia-4": {
    "kind": "model",
    "glossary": {
      "one_liner": "闭环模拟：驾驶系统的动作会影响模拟器下一步返回的观察。"
    },
    "input": {
      "value": "记录的传感器数据。",
      "tags": [
        "sensor"
      ]
    },
    "output": {
      "value": "相机与雷达的多模态模拟数据。",
      "tags": [
        "visual",
        "spatial",
        "sensor"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "driving_action"
      ],
      "domains": [
        "driving"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "driving_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "driving": [
            "profile:gaia-4-context"
          ]
        }
      },
      "reviewed_on": "2026-10-01"
    }
  },
  "dreamerv3": {
    "kind": "model",
    "input": {
      "value": "感官观测及动作。",
      "tags": [
        "visual",
        "sensor",
        "action"
      ]
    },
    "output": {
      "value": "未来内部表征和奖励。",
      "tags": [
        "latent"
      ]
    },
    "glossary": {},
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "games",
        "robotics"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "games": [
            "profile:dreamerv3-source-1"
          ],
          "robotics": [
            "profile:dreamerv3-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "锁定 Nature 2025 的第三代算法。机器人与游戏标签对应论文的模拟任务范围；论文结果和作者维护的重实现分别理解。"
    }
  },
  "dino-wm": {
    "kind": "model",
    "input": {
      "value": "视觉表征和动作。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "未来视觉 patch 特征。",
      "tags": [
        "latent"
      ]
    },
    "glossary": {
      "input": "完整流程先从图像提取 DINOv2 patch 特征；起点和目标图像用于视觉目标规划。",
      "output": "特征预测不等于直接生成可观看的视频。"
    },
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "robotics": [
            "profile:dino-wm-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "按 ICML 2025 论文对象整理，2024 是首稿年份。机器人标签指所列模拟控制与操作任务，不表示通用真机部署；DINOv2 编码器与 DINO-WM 分开。"
    }
  },
  "daydreamer": {
    "kind": "model",
    "input": {
      "value": "本体感知和视觉观测。",
      "tags": [
        "sensor",
        "visual"
      ]
    },
    "output": {
      "value": "动作条件下预测的内部表征序列。",
      "tags": [
        "latent"
      ]
    },
    "glossary": {
      "input": "动作条件另见“动作的作用”；各机器人平台的具体观测组成需要分别查阅论文。",
      "output": "作者实现还从内部状态与表征重建输入，预测奖励和回合结束；这里先列核心序列预测对象。"
    },
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "robot_action"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "robot_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "robotics": [
            "profile:daydreamer-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "区分不同机器人平台的观测、动作与奖励设置。想象式策略训练发生在内部表征空间；本页不将四个平台的实验概括为同一套通用权重。"
    }
  },
  "diamond-atari": {
    "kind": "model",
    "input": {
      "value": "历史游戏帧和智能体动作。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "下一游戏画面帧。",
      "tags": [
        "visual"
      ]
    },
    "glossary": {},
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "games"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "games": [
            "profile:diamond-atari-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "Atari 论文对象与单独训练的 CS:GO 演示分开。生成画面不是游戏程序执行的真实状态；不同训练数据、模型规模和分支不合并。"
    }
  },
  "ctrl-world": {
    "kind": "model",
    "input": {
      "value": "初始画面和动作序列。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "联合预测的多视角视频，包括腕部视角。",
      "tags": [
        "visual"
      ]
    },
    "glossary": {},
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "robot_action"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "robot_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "robotics": [
            "profile:ctrl-world-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "对应 ICLR 2026 原项目，论文索引所读版本为 2026-03-01 v3。Ctrl-World、所用 VLA 策略和后续 VLAW 流程分别记录，不把评估结果外推到任意机器人。"
    }
  },
  "wham-2025": {
    "kind": "model",
    "input": {
      "value": "游戏图像和控制器动作条件。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "预测的游戏画面和控制器动作序列，依具体模式而定。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "glossary": {},
    "classification": {
      "types": [
        "latent",
        "simulation"
      ],
      "controls": [
        "agent_action",
        "scene_edit"
      ],
      "domains": [
        "games"
      ],
      "basis": {
        "types": {
          "latent": [
            "profile:wham-2025-source-2"
          ],
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ],
          "scene_edit": [
            "profile:wham-2025-source-2"
          ]
        },
        "domains": {
          "games": [
            "profile:wham-2025-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "WHAM 的核心自回归预测对象是观测与动作 token，再解码为可观察序列。图像编辑条件不等于三维场景输出；WHAM-RT、Muse 研究计划与首个论文模型分开。"
    }
  },
  "vista-2024": {
    "kind": "model",
    "input": {
      "value": "历史帧与高层意图，包括指令和目标点。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "未来驾驶场景的视觉预测。",
      "tags": [
        "visual"
      ]
    },
    "glossary": {
      "input": "其他配置还支持转向角、速度与轨迹等控制；按所用配置分别查阅。"
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "driving_action"
      ],
      "domains": [
        "driving"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "driving_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "driving": [
            "profile:vista-2024-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "限定 OpenDriveLab Vista 2024 对象，所读论文索引为 v5。不同控制配置并非同时必需，也不等于完整闭环驾驶系统；不与 GenAD 或其他同名项目混用。"
    }
  },
  "atlas": {
    "kind": "model",
    "input": {
      "value": "文本、图像、相机位姿和三维深度图；视频表示为图像序列。",
      "tags": [
        "text",
        "visual",
        "spatial"
      ]
    },
    "output": {
      "value": "图像帧与三维深度信息，可组成点云或 Gaussian splats。",
      "tags": [
        "visual",
        "spatial"
      ]
    },
    "glossary": {},
    "classification": {
      "types": [
        "simulation",
        "spatial"
      ],
      "controls": [
        "language",
        "camera"
      ],
      "domains": [
        "spatial_creation",
        "robotics"
      ],
      "basis": {
        "types": {
          "simulation": [
            "profile:atlas-source-1"
          ],
          "spatial": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "technical:input"
          ],
          "camera": [
            "technical:input"
          ]
        },
        "domains": {
          "spatial_creation": [
            "profile:atlas-source-1"
          ],
          "robotics": [
            "profile:atlas-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "依据 2026-09-01 官方介绍对象。视频模拟与可用三维场景输出分别打标签；相机控制不自动推出逐步机器人动作接口，动作条件暂保留未知。"
    }
  },
  "leworldmodel": {
    "kind": "model",
    "input": {
      "value": "当前图像的潜在表征和动作。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "下一帧观测的潜在表征。",
      "tags": [
        "latent"
      ]
    },
    "glossary": {
      "input": "编码器把观测图像转换为表征；起点和目标图像由规划模块使用。"
    },
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "robotics": [
            "profile:leworldmodel-source-1"
          ]
        }
      },
      "reviewed_on": "2026-10-01",
      "scope_note": "依据保存的 2026 项目页与作者仓库基础说明，不将 v1/v3 差异视为已完成复核。机器人领域对应所列模拟操作任务；可视化视频不等于模型直接生成视频。"
    }
  },
  "h-jepa": {
    "kind": "model",
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "robotics": [
            "profile:h-jepa-paper"
          ]
        }
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "层次潜在预测；图像解码用于可视化。机器人领域包含论文的操作任务与 DROID 离线评估，不表示已验证实体闭环控制。"
    },
    "input": {
      "value": "过去的潜在状态与动作。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "各层的未来潜在状态。",
      "tags": [
        "latent"
      ]
    }
  },
  "uniwam": {
    "kind": "model",
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "language"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "profile:uniwam-paper-v3"
          ]
        },
        "domains": {
          "robotics": [
            "profile:uniwam-code"
          ]
        }
      },
      "reviewed_on": "2026-10-10",
      "scope_note": "预测未来视觉并生成动作；视觉生成器内部使用潜在空间，不据此归为表征与状态预测。外部动作条件推演接口未确认。"
    },
    "input": {
      "value": "当前观测、本体感知状态与任务指令。",
      "tags": [
        "visual",
        "sensor",
        "text"
      ]
    },
    "output": {
      "value": "语言输出、未来观测与动作。",
      "tags": [
        "text",
        "visual",
        "action"
      ]
    }
  },
  "deltaworld": {
    "kind": "model",
    "classification": {
      "types": [
        "latent",
        "simulation"
      ],
      "controls": [
        "robot_action"
      ],
      "domains": [
        "robotics"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:one_liner"
          ],
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "robot_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "robotics": [
            "profile:deltaworld-paper"
          ]
        }
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "核心预测目标是动作引起的潜在状态增量；同时解码未来 RGB，故两种模型类型并列。范围为论文的机器人操作任务。"
    },
    "input": {
      "value": "初始视觉上下文与机器人动作序列。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "解码后的未来 RGB 观测。",
      "tags": [
        "visual"
      ]
    }
  },
  "v2x-wam": {
    "kind": "model",
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "driving_action"
      ],
      "domains": [
        "driving"
      ],
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "driving_action": [
            "technical:action_conditioned"
          ]
        },
        "domains": {
          "driving": [
            "profile:v2x-wam-paper"
          ]
        }
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "预测结构化交通状态（占据与动态流），按车辆轨迹和动作条件展开；不是可复用 3D 场景资产或通用视频模型。"
    },
    "input": {
      "value": "车路协同场景表征、所选轨迹与动作序列。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "未来交通占据与动态流。",
      "tags": [
        "latent"
      ]
    }
  },
  "gwm-worlds-2": {
    "kind": "model",
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "language",
        "camera",
        "navigation"
      ],
      "domains": [
        "games",
        "robotics"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "technical:input"
          ],
          "camera": [
            "technical:input"
          ],
          "navigation": [
            "profile:gwm-worlds-2-release"
          ]
        },
        "domains": {
          "games": [
            "profile:gwm-worlds-2-release"
          ],
          "robotics": [
            "profile:gwm-worlds-2-release"
          ]
        }
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "研究预览中的实时视听模拟；游戏、虚拟人物与机器人模拟用途不表示各领域均已完成外部验证。视频预填充描述存在冲突。"
    },
    "input": {
      "value": "场景初始提示与首帧、当前相机和文本动作、历史生成帧。",
      "tags": [
        "text",
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "逐步生成的视频与音频。",
      "tags": [
        "visual",
        "audio"
      ]
    }
  },
  "alayaworld-v1-1": {
    "kind": "model",
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "camera"
      ],
      "domains": [
        "unknown"
      ],
      "basis": {
        "types": {
          "simulation": [
            "profile:alayaworld-v1-1-paper"
          ]
        },
        "controls": {
          "camera": [
            "technical:input"
          ]
        },
        "domains": {}
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "v1.1 的 ViGeo 空间记忆条件视频路径；旧 DA3、共享 Stage1 与 v1.1 AR/DMD 分别看待。应用领域与外部动作条件接口未确认。"
    },
    "input": {
      "value": "经重新渲染的空间记忆与运动感知视觉条件。",
      "tags": [
        "visual",
        "spatial"
      ]
    },
    "output": {
      "value": "按视角条件生成的视频。",
      "tags": [
        "visual"
      ]
    }
  },
  "alaya-evoke": {
    "kind": "model",
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "camera",
        "language"
      ],
      "domains": [
        "unknown"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "camera": [
            "technical:input"
          ],
          "language": [
            "technical:input"
          ]
        },
        "domains": {}
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "Evoke 基础模型与 Turbo 同页，二者不与 AlayaWorld v1.1 互相替代。相机与提示控制不单独证明智能体动作条件动力学。"
    },
    "input": {
      "value": "时间变化的文本、相机轨迹与投影历史。",
      "tags": [
        "text",
        "spatial",
        "visual"
      ]
    },
    "output": {
      "value": "交互生成的视频。",
      "tags": [
        "visual"
      ]
    }
  },
  "inspatio-world-1-5": {
    "kind": "model",
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "camera"
      ],
      "domains": [
        "unknown"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "camera": [
            "profile:inspatio-world-1-5-code"
          ]
        },
        "domains": {}
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "仅采用 v1.5 的公开推理说明。模型卡、完整训练报告与权重许可仍缺；不由旧版资料或深度后端外推本版性能与 3D 输出。"
    },
    "input": {
      "value": "单张图像、四张图像或视频，以及用户提供的目标相机轨迹。",
      "tags": [
        "visual",
        "spatial"
      ]
    },
    "output": {
      "value": "源画面、渲染画面、遮罩与预测视频文件。",
      "tags": [
        "visual"
      ]
    }
  },
  "solaris": {
    "kind": "model",
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "agent_action",
        "language"
      ],
      "domains": [
        "web"
      ],
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:action_conditioned"
          ],
          "language": [
            "profile:solaris-release"
          ]
        },
        "domains": {
          "web": [
            "profile:solaris-release"
          ]
        }
      },
      "reviewed_on": "2026-10-07",
      "scope_note": "生成界面像素的接口世界模型；用户点击与拖拽作为下一帧条件，语言模型负责界面演进，不据演示推断真实后端已运行。"
    },
    "input": {
      "value": "用户点击、拖拽等已发生的交互。",
      "tags": [
        "action"
      ]
    },
    "output": {
      "value": "逐帧生成的视觉界面。",
      "tags": [
        "visual"
      ]
    }
  },
  "odyssey-3": {
    "kind": "family",
    "glossary": {},
    "input": {
      "value": "历史观察、文本提示和动作输入；预览支持第一/第三人称导航与独立相机移动。",
      "tags": [
        "visual",
        "text",
        "action"
      ]
    },
    "output": {
      "value": "随交互持续生成的未来环境视频。",
      "tags": [
        "visual"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "language",
        "camera",
        "navigation"
      ],
      "domains": [
        "physical_ai"
      ],
      "reviewed_on": "2026-10-10",
      "scope_note": "本页收录 Odyssey-3 家族。视频输出归为视频与传感模拟；物理 AI 标签表示官方用途，不表示对物理准确性的独立认证。",
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "profile:odyssey-3-release"
          ],
          "camera": [
            "profile:odyssey-3-release"
          ],
          "navigation": [
            "profile:odyssey-3-release"
          ]
        },
        "domains": {
          "physical_ai": [
            "profile:odyssey-3-release"
          ]
        }
      }
    }
  },
  "dreamtrue": {
    "kind": "model",
    "glossary": {},
    "input": {
      "value": "初始场景与机器人动作条件；轨迹经几何校准后渲染到图像空间。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "output": {
      "value": "多视角未来机器人操作视频。",
      "tags": [
        "visual"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "robot_action"
      ],
      "domains": [
        "robotics"
      ],
      "reviewed_on": "2026-10-10",
      "scope_note": "论文 v1 的多视角动作条件视频预测；代码公开不自动代表权重或完整数据已开放。",
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "robot_action": [
            "technical:input"
          ]
        },
        "domains": {
          "robotics": [
            "profile:dreamtrue-paper"
          ]
        }
      }
    }
  },
  "planwam": {
    "kind": "model",
    "glossary": {
      "output": "潜在表征是模型内部编码；这里用它帮助轨迹生成和选择。"
    },
    "input": {
      "value": "历史多视角相机观察、自车状态与导航信息。",
      "tags": [
        "visual"
      ]
    },
    "output": {
      "value": "面向规划的未来潜在表征，以及由其引导生成和选择的自车轨迹。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "unknown"
      ],
      "domains": [
        "driving"
      ],
      "reviewed_on": "2026-10-10",
      "scope_note": "论文 v1；核心世界模型预测未来潜在表征，不生成可复用三维场景。导航信息不自动视为外部动作条件接口。",
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {},
        "domains": {
          "driving": [
            "profile:planwam-paper"
          ]
        }
      }
    }
  },
  "long-wam": {
    "kind": "family",
    "glossary": {},
    "input": {
      "value": "观测视频历史、机器人状态与语言任务指令；具体格式随变体而定。",
      "tags": [
        "visual",
        "text"
      ]
    },
    "output": {
      "value": "未来视频潜变量与机器人动作块；IDM 先预测视频，再生成动作。",
      "tags": [
        "visual",
        "action"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "language"
      ],
      "domains": [
        "robotics"
      ],
      "reviewed_on": "2026-10-10",
      "scope_note": "Long-WAM 家族包含 IDM 与 CoD 等变体；本页的权重状态由 LIBERO-IDM 等公开分发材料支持，不表示任意机器人配置即装即用。",
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "technical:input"
          ]
        },
        "domains": {
          "robotics": [
            "profile:long-wam-paper"
          ]
        }
      }
    }
  },
  "ppwm": {
    "kind": "model",
    "glossary": {},
    "input": {
      "value": "近期潜在状态与已执行动作组成的上下文，以及待预测的动作序列。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "有限跨度内的多步未来潜在表征轨迹。",
      "tags": [
        "latent"
      ]
    },
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "agent_action"
      ],
      "domains": [
        "unknown"
      ],
      "reviewed_on": "2026-10-10",
      "scope_note": "论文 v1 的并行预测方法及实例；控制标签由动作序列条件支持，应用领域暂未知。",
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "agent_action": [
            "technical:input"
          ]
        },
        "domains": {}
      }
    }
  },
  "token-world": {
    "kind": "model",
    "glossary": {},
    "input": {
      "value": "压缩的 VLM 视觉 token、本体感知与机器人动作历史。",
      "tags": [
        "latent",
        "action"
      ]
    },
    "output": {
      "value": "未来压缩 token 状态，可映射回策略使用的 VLM 视觉表征。",
      "tags": [
        "latent"
      ]
    },
    "classification": {
      "types": [
        "latent"
      ],
      "controls": [
        "robot_action"
      ],
      "domains": [
        "robotics"
      ],
      "reviewed_on": "2026-10-10",
      "scope_note": "论文 v2 的 VLM token 动力学模型；RGB 输出仅用于可视化，不因此增加视频与传感模拟标签。",
      "basis": {
        "types": {
          "latent": [
            "technical:output"
          ]
        },
        "controls": {
          "robot_action": [
            "technical:input"
          ]
        },
        "domains": {
          "robotics": [
            "profile:token-world-paper"
          ]
        }
      }
    }
  },
  "pixverse-r2": {
    "kind": "model",
    "glossary": {},
    "input": {
      "value": "历史环境状态，以及实时文本、参考材料、音频、WASD 等动作输入。",
      "tags": [
        "visual",
        "text",
        "audio",
        "action"
      ]
    },
    "output": {
      "value": "连续更新的同步视频与音频片段。",
      "tags": [
        "visual",
        "audio"
      ]
    },
    "classification": {
      "types": [
        "simulation"
      ],
      "controls": [
        "language",
        "navigation",
        "camera",
        "audio"
      ],
      "domains": [
        "games"
      ],
      "reviewed_on": "2026-10-10",
      "scope_note": "R2 实时音视频模型；导航和相机标签同时参考官方产品说明，不采用第三方排名。",
      "basis": {
        "types": {
          "simulation": [
            "technical:output"
          ]
        },
        "controls": {
          "language": [
            "profile:pixverse-r2-release"
          ],
          "navigation": [
            "profile:pixverse-r2-product"
          ],
          "camera": [
            "profile:pixverse-r2-product"
          ],
          "audio": [
            "profile:pixverse-r2-release"
          ]
        },
        "domains": {
          "games": [
            "profile:pixverse-r2-product"
          ]
        }
      }
    }
  }
}