{
  "schema_version": "paper_public_manifest_v1",
  "paper_id": "chunktrust_2026_10",
  "slug": "chunktrust",
  "title": "Want VLA to be more reliable? Is the cost more replanning?",
  "authors": [],
  "source": {
    "arxiv_id": "2609.39754",
    "pdf_url": "https://arxiv.org/pdf/2609.39754",
    "project_url": "",
    "github_url": "",
    "huggingface_url": "",
    "original_source": "https://arxiv.org/pdf/2609.39754"
  },
  "site": {
    "post_url": "/posts/chunktrust",
    "canonical_url": "https://haiguangboy.com/posts/chunktrust",
    "cover_image": "https://static.haiguangboy.com/papers/chunktrust/cover.webp"
  },
  "taxonomy": {
    "domain": "embodied_ai",
    "track": "vla",
    "tasks": [
      "embodied_ai",
      "vla",
      "action_generation",
      "robotics",
      "state_prediction",
      "Embodied Intelligence",
      "Robotics",
      "wam",
      "chunktrust"
    ],
    "related_topics": [
      {
        "paper_id": "smoothrl_online_reinforcement_learning_during_asynchronous_execution_2026_09",
        "title": "SmoothRL: Only account for actions that are actually executed",
        "url": "https://haiguangboy.com/posts/smoothrl_online_reinforcement_learning_during_asynchronous_execution",
        "relation": "contrast",
        "summary": "SmoothRL: Only account for actions that are actually executed",
        "strength": "strong"
      },
      {
        "paper_id": "agile_wam_an_agile_tactile_world_action_model_for_contact_rich_robot_control_2026_09",
        "title": "Agile-WAM: Lightweight visuotactile joint prediction",
        "url": "https://haiguangboy.com/posts/agile-wam",
        "relation": "same_track",
        "summary": "Evaluation scope and cost accounting must be stated separately; validates that small-sample and best-of-n evaluations cannot be extrapolated to stable general capabilities",
        "strength": "strong"
      },
      {
        "paper_id": "dextouch_wm_learning_action_conditioned_tactile_world_models_from_human_touch_fo_2026_09",
        "title": "DexTouch-WM: Human tactile pretraining",
        "url": "https://haiguangboy.com/posts/dextouch-wm",
        "relation": "same_track",
        "summary": "Internal quality and Beta posterior are not calibrated success probabilities; validates that correlation does not equal success rate, nor absolute score calibration",
        "strength": "strong"
      },
      {
        "paper_id": "hyperworld_hypergraph_structured_state_serialization_improves_learned_textual_wo_2026_09",
        "title": "HyperWorld: Take it apart, generalize away 31 points",
        "url": "https://haiguangboy.com/posts/hyperworld_hypergraph_structured_state_serialization_improves_learned_textual_wo",
        "relation": "same_track",
        "summary": "HyperWorld: Take it apart, generalize away 31 points",
        "strength": "strong"
      },
      {
        "paper_id": "google_blog_gemini_robotics_er_2_20260730_2026_07",
        "title": "Introducing Gemini Robotics ER 2",
        "url": "https://haiguangboy.com/posts/google_blog_gemini_robotics_er_2_20260730",
        "relation": "same_track",
        "summary": "Average improvement cannot hide regressions, nor can it claim to comprehensively outperform other selectors; validates that the numbers lack controls: no named competitors, no baseline values, and comparison charts that cannot be verified",
        "strength": "strong"
      },
      {
        "paper_id": "openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra_2026_09",
        "title": "OpenWAM: The right answer changes with scale",
        "url": "https://haiguangboy.com/posts/openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra",
        "relation": "same_track",
        "summary": "OpenWAM: The right answer changes with scale",
        "strength": "strong"
      }
    ]
  },
  "analyst_take": {
    "type": "author_opinion",
    "text": "💡 Put it in the technical lineage of the execution layer\nLegato improves action chunk transitions through training; SmoothRL learns residual corrections to executed actions; ChunkTrust preserves action values and adjusts when to reacquire feedback.\n\nThe three address different stages. To judge whether an execution scheme is worth deploying, success rate, feedback latency, model calls, and total computation should be considered together, rather than reporting only the selector's millisecond overhead."
  },
  "ruling": {
    "importance_score": 3.0,
    "one_sentence": "Want VLA to be more reliable? Is the cost more replanning?"
  },
  "asset_base_url": "https://static.haiguangboy.com/papers/chunktrust",
  "assets": [
    {
      "type": "cover_image",
      "object_key": "papers/chunktrust/cover.webp",
      "url": "https://static.haiguangboy.com/papers/chunktrust/cover.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "post_cover"
    },
    {
      "type": "pdf_screenshot",
      "object_key": "papers/chunktrust/page_01.webp",
      "url": "https://static.haiguangboy.com/papers/chunktrust/page_01.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "paper_first_page"
    },
    {
      "type": "pdf_screenshot",
      "object_key": "papers/chunktrust/key_figure.webp",
      "url": "https://static.haiguangboy.com/papers/chunktrust/key_figure.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "method_figure"
    },
    {
      "type": "public_manifest",
      "object_key": "papers/chunktrust/public_manifest.json",
      "url": "https://static.haiguangboy.com/papers/chunktrust/public_manifest.json",
      "content_type": "application/json; charset=utf-8",
      "upload_status": "pending",
      "role": "public_manifest"
    },
    {
      "type": "public_brief",
      "object_key": "papers/chunktrust/public_brief.md",
      "url": "https://static.haiguangboy.com/papers/chunktrust/public_brief.md",
      "content_type": "text/markdown; charset=utf-8",
      "upload_status": "pending",
      "role": "public_brief"
    }
  ],
  "published_at": "2026-10-14T17:08:59+08:00",
  "created_at": "2026-10-14T17:08:59+08:00",
  "updated_at": "2026-10-14T17:08:59+08:00"
}