{
  "schema_version": "paper_public_manifest_v1",
  "paper_id": "mt-wam_2026_09",
  "slug": "mt-wam",
  "title": "MT-WAM：未来监督如何服务动作",
  "authors": [],
  "source": {
    "arxiv_id": "2609.21474",
    "pdf_url": "https://arxiv.org/pdf/2609.21474",
    "project_url": "",
    "github_url": "",
    "huggingface_url": "",
    "original_source": "https://arxiv.org/abs/2609.21474"
  },
  "site": {
    "post_url": "/posts/mt-wam",
    "canonical_url": "https://haiguangboy.com/posts/mt-wam",
    "cover_image": "https://static.haiguangboy.com/papers/mt-wam/cover.webp"
  },
  "taxonomy": {
    "domain": "embodied_ai",
    "track": "world_model",
    "tasks": [
      "embodied_ai",
      "world_model",
      "action_generation",
      "robotics",
      "state_prediction",
      "具身智能",
      "机器人",
      "世界模型",
      "mtwam",
      "fastwam",
      "表征学习"
    ],
    "related_topics": [
      {
        "paper_id": "openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra_2026_09",
        "title": "OpenWAM:正确答案会随规模变",
        "url": "https://haiguangboy.com/posts/openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra",
        "relation": "contrast",
        "summary": "OpenWAM:正确答案会随规模变",
        "strength": "strong"
      },
      {
        "paper_id": "orca_2026_07",
        "title": "π0.5抓勺子失败就原地哆嗦,Orca却靠刷视频悟出的物理直觉更进一步",
        "url": "https://haiguangboy.com/posts/orca",
        "relation": "contrast",
        "summary": "世界模型的关键是可读出的状态",
        "strength": "strong"
      },
      {
        "paper_id": "omega_0_a_latent_predictive_world_action_model_for_concurrent_humanoid_loco_mani_2026_08",
        "title": "未来预测越准，机器人反而越迟滞",
        "url": "https://haiguangboy.com/posts/omega-0",
        "relation": "contrast",
        "summary": "复制视频主干末端形成双流分支，而非只加输出头 contradicts 核心设计:未来预测不当视频生成目标,而是重建无关的紧凑预测信号,动作分支不需要像素级生成或测试时视频转动作",
        "strength": "strong"
      },
      {
        "paper_id": "slim_05b_learning_action_grounded_predictive_latents_for_robot_manipulation_2026_08",
        "title": "预测误差更低，表征却已经塌了",
        "url": "https://haiguangboy.com/posts/slim-05b",
        "relation": "contrast",
        "summary": "保留FAST-WAM直接动作接口，增加两类未来监督 contradicts Stage 2 把未来潜变量整个撤掉,只留被 Stage 1 塑形过的预测性 slot",
        "strength": "strong"
      },
      {
        "paper_id": "agile_wam_an_agile_tactile_world_action_model_for_contact_rich_robot_control_2026_09",
        "title": "Agile-WAM：轻量视触觉联合预测",
        "url": "https://haiguangboy.com/posts/agile-wam",
        "relation": "same_track",
        "summary": "保留FAST-WAM直接动作接口，增加两类未来监督 validates 作者主张：输入哪些模态，也应学习这些模态的未来",
        "strength": "strong"
      },
      {
        "paper_id": "jepa_wam_learning_vision_language_action_policies_with_joint_embedding_world_modeling_2026_08",
        "title": "JEPA-WAM：创新方法让世界模型更懂状态变化",
        "url": "https://haiguangboy.com/posts/jepa-wam",
        "relation": "same_track",
        "summary": "作者判断：省掉未来生成后，应优先改善当前一次前向表征 validates 核心主张:转移监督必须直接塑造生成动作的那个主干,外挂和当上下文都不行",
        "strength": "strong"
      }
    ]
  },
  "analyst_take": {
    "type": "author_opinion",
    "text": "💡 放到相关工作里看\nTacWAM把未来触觉用于监督，并在嵌套消融中发现，逐步放开信息交互会损害控制；这与MT-WAM的非对称分工呼应。\n\n但OpenWAM发现，切断动作对世界特征的访问会掉约5分。两者并非直接互相否定：世界流与辅助视觉特征流不是同一种信号，MT-WAM也保留了动作对运动流的读取。工程上真正该逐项验证的是：哪个目标改善表征，哪条信息通路改善控制。"
  },
  "ruling": {
    "importance_score": 3.0,
    "one_sentence": "MT-WAM：未来监督如何服务动作"
  },
  "asset_base_url": "https://static.haiguangboy.com/papers/mt-wam",
  "assets": [
    {
      "type": "pdf_screenshot",
      "object_key": "papers/mt-wam/page_01.webp",
      "content_type": "image/webp",
      "upload_status": "uploaded",
      "bucket": "paper-assets",
      "url": "https://static.haiguangboy.com/papers/mt-wam/page_01.webp",
      "role": "paper_first_page",
      "size_bytes": 284928
    },
    {
      "type": "pdf_screenshot",
      "object_key": "papers/mt-wam/key_figure.webp",
      "content_type": "image/webp",
      "upload_status": "uploaded",
      "bucket": "paper-assets",
      "url": "https://static.haiguangboy.com/papers/mt-wam/key_figure.webp",
      "role": "method_figure",
      "size_bytes": 232942
    },
    {
      "type": "cover_image",
      "object_key": "papers/mt-wam/cover.webp",
      "content_type": "image/webp",
      "upload_status": "uploaded",
      "bucket": "paper-assets",
      "url": "https://static.haiguangboy.com/papers/mt-wam/cover.webp",
      "role": "post_cover",
      "size_bytes": 55518
    },
    {
      "type": "public_brief",
      "object_key": "papers/mt-wam/public_brief.md",
      "content_type": "text/markdown; charset=utf-8",
      "upload_status": "uploaded",
      "bucket": "paper-assets",
      "url": "https://static.haiguangboy.com/papers/mt-wam/public_brief.md",
      "role": "public_brief",
      "size_bytes": 3822
    },
    {
      "type": "public_brief",
      "object_key": "papers/mt-wam/public_brief.en.md",
      "content_type": "text/markdown; charset=utf-8",
      "upload_status": "uploaded",
      "bucket": "paper-assets",
      "url": "https://static.haiguangboy.com/papers/mt-wam/public_brief.en.md",
      "role": "public_brief_en",
      "size_bytes": 4882
    },
    {
      "type": "public_manifest",
      "object_key": "papers/mt-wam/public_manifest.en.json",
      "content_type": "application/json; charset=utf-8",
      "upload_status": "uploaded",
      "bucket": "paper-assets",
      "url": "https://static.haiguangboy.com/papers/mt-wam/public_manifest.en.json",
      "role": "public_manifest_en",
      "size_bytes": 6852
    },
    {
      "type": "public_manifest",
      "object_key": "papers/mt-wam/public_manifest.json",
      "content_type": "application/json; charset=utf-8",
      "upload_status": "uploaded",
      "bucket": "paper-assets",
      "url": "https://static.haiguangboy.com/papers/mt-wam/public_manifest.json",
      "role": "public_manifest",
      "size_bytes": 7119
    }
  ],
  "published_at": "2026-09-28T11:43:53+08:00",
  "created_at": "2026-09-28T11:43:53+08:00",
  "updated_at": "2026-09-28T11:44:08+08:00"
}
