{
  "schema_version": "paper_public_manifest_v1",
  "paper_id": "vlaw_2026_09",
  "slug": "vlaw",
  "title": "VLAW: Policy and World Model Improve Together",
  "authors": [],
  "source": {
    "arxiv_id": "2602.12063",
    "pdf_url": "https://arxiv.org/pdf/2602.12063",
    "project_url": "",
    "github_url": "",
    "huggingface_url": "",
    "original_source": "https://arxiv.org/pdf/2602.12063"
  },
  "site": {
    "post_url": "/posts/vlaw",
    "canonical_url": "https://haiguangboy.com/posts/vlaw",
    "cover_image": "https://static.haiguangboy.com/papers/vlaw/cover.webp"
  },
  "taxonomy": {
    "domain": "embodied_ai",
    "track": "world_model",
    "tasks": [
      "embodied_ai",
      "world_model",
      "vla",
      "action_generation",
      "robotics",
      "state_prediction",
      "Embodied Intelligence",
      "Robotics",
      "World Models",
      "vlaw",
      "Robot Learning"
    ],
    "related_topics": [
      {
        "paper_id": "hydra_0_action_flow_for_generalist_world_modeling_and_control_2026_08",
        "title": "NVIDIA's Motion Flow: A New Standard for Robot Actions",
        "url": "https://haiguangboy.com/posts/hydra-0-action-flow",
        "relation": "contrast",
        "summary": "Still relies on real-robot interaction, manual resetting, and success labels contradicts the redefinition of the modeling target: not reproducing task-specific behaviors, but modeling the consequences of robot motion",
        "strength": "strong"
      },
      {
        "paper_id": "pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation_2026_08",
        "title": "PointWorld: Point Flow Unifies State and Action",
        "url": "https://haiguangboy.com/posts/pointworld_scaling_3d_world_models_for_in_the_wild_robotic_manipulation",
        "relation": "same_track",
        "summary": "PointWorld: Point Flow Unifies State and Action",
        "strength": "strong"
      },
      {
        "paper_id": "rl_100_performant_robotic_manipulation_with_real_world_reinforcement_learning_2026_08",
        "title": "RL Should Not Learn from Scratch, It Should Do Post-Training",
        "url": "https://haiguangboy.com/posts/rl-100",
        "relation": "same_track",
        "summary": "Still relies on real-robot interaction, manual resetting, and success labels validates that autonomous resetting remains an unresolved practical bottleneck, which the authors themselves list as future work",
        "strength": "strong"
      },
      {
        "paper_id": "skild_blog_introducing_s1_in_context_learning_20260817_2026_08",
        "title": "S1: Robotics' GPT-3 Emergence Moment",
        "url": "https://haiguangboy.com/posts/skild_blog_introducing_s1_in_context_learning_20260817",
        "relation": "same_track",
        "summary": "S1: Robotics' GPT-3 Emergence Moment",
        "strength": "strong"
      },
      {
        "paper_id": "towards_generalizable_visually_grounded_exploration_of_household_devices_2026_09",
        "title": "VGEBench: Generalizable Visual Exploration",
        "url": "https://haiguangboy.com/posts/towards_generalizable_visually_grounded_exploration_of_household_devices",
        "relation": "same_track",
        "summary": "VGEBench: Generalizable Visual Exploration",
        "strength": "strong"
      },
      {
        "paper_id": "robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode_2026_07",
        "title": "It's Not About \"Whether We Need a World Model,\" but \"Whether What It Outputs Has Structure\"",
        "url": "https://haiguangboy.com/posts/robointer15_a_holistic_intermediate_representation_suite_for_embodied_world_mode",
        "relation": "same_track",
        "summary": "It's Not About \"Whether We Need a World Model,\" but \"Whether What It Outputs Has Structure\"",
        "strength": "strong"
      }
    ]
  },
  "analyst_take": {
    "type": "author_opinion",
    "text": "💡 Read This Alongside the WAM Survey\nThe survey reminds us: world models pursue prediction accuracy, policies pursue returns, and the two have different objectives. VLAW grounds this division of labor in the data: the world model learns from both successes and failures, while the policy learns from filtered successful experiences. They improve together, without requiring both to learn from the same set of trajectories.\n\nLooking back in six months: under the same real-robot budget, as tasks and iteration rounds scale up, can this alternating improvement continue to yield gains?"
  },
  "ruling": {
    "importance_score": 3.0,
    "one_sentence": "VLAW: Policy and World Model Improve Together"
  },
  "asset_base_url": "https://static.haiguangboy.com/papers/vlaw",
  "assets": [
    {
      "type": "cover_image",
      "object_key": "papers/vlaw/cover.webp",
      "url": "https://static.haiguangboy.com/papers/vlaw/cover.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "post_cover"
    },
    {
      "type": "pdf_screenshot",
      "object_key": "papers/vlaw/page_01.webp",
      "url": "https://static.haiguangboy.com/papers/vlaw/page_01.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "paper_first_page"
    },
    {
      "type": "pdf_screenshot",
      "object_key": "papers/vlaw/key_figure.webp",
      "url": "https://static.haiguangboy.com/papers/vlaw/key_figure.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "method_figure"
    },
    {
      "type": "public_manifest",
      "object_key": "papers/vlaw/public_manifest.json",
      "url": "https://static.haiguangboy.com/papers/vlaw/public_manifest.json",
      "content_type": "application/json; charset=utf-8",
      "upload_status": "pending",
      "role": "public_manifest"
    },
    {
      "type": "public_brief",
      "object_key": "papers/vlaw/public_brief.md",
      "url": "https://static.haiguangboy.com/papers/vlaw/public_brief.md",
      "content_type": "text/markdown; charset=utf-8",
      "upload_status": "pending",
      "role": "public_brief"
    }
  ],
  "published_at": "2026-09-19T08:32:49+08:00",
  "created_at": "2026-09-19T08:32:49+08:00",
  "updated_at": "2026-09-19T08:32:49+08:00"
}