{
  "schema_version": "paper_public_manifest_v1",
  "paper_id": "koda_2026_10",
  "slug": "koda",
  "title": "KODA: Converting Human Demonstrations into Humanoid Whole-Body Data",
  "authors": [],
  "source": {
    "arxiv_id": "2610.21152",
    "pdf_url": "https://arxiv.org/pdf/2610.21152",
    "project_url": "",
    "github_url": "",
    "huggingface_url": "",
    "original_source": "https://arxiv.org/abs/2610.21152"
  },
  "site": {
    "post_url": "/posts/koda",
    "canonical_url": "https://haiguangboy.com/posts/koda",
    "cover_image": "https://static.haiguangboy.com/papers/koda/cover.webp"
  },
  "taxonomy": {
    "domain": "embodied_ai",
    "track": "vla",
    "tasks": [
      "embodied_ai",
      "vla",
      "action_generation",
      "robotics",
      "state_prediction",
      "Embodied Intelligence",
      "Humanoid Robots",
      "Human Demonstrations",
      "Whole-Body Control",
      "gr00t"
    ],
    "related_topics": [
      {
        "paper_id": "an_open_foundation_model_towards_2026_07",
        "title": "An_Open_Foundation_Model_Towards",
        "url": "https://haiguangboy.com/posts/an_open_foundation_model_towards",
        "relation": "contrast",
        "summary": "Real-robot training uses co-training of robot and converted human data, which contradicts the claim that joint co-training of heterogeneous human-robot data is a structurally suboptimal solution",
        "strength": "strong"
      },
      {
        "paper_id": "openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra_2026_09",
        "title": "OpenWAM: The Correct Answer Changes with Scale",
        "url": "https://haiguangboy.com/posts/openwam_an_open_modular_exploration_towards_systematic_world_action_model_pretra",
        "relation": "contrast",
        "summary": "OpenWAM: The Correct Answer Changes with Scale",
        "strength": "strong"
      },
      {
        "paper_id": "transformer-transformer_blog_transformer_transformer_a_unified_20260807_2026_08",
        "title": "Transformer Transformer: A Unified Model for Motion-Conditioned Robot Co-design",
        "url": "https://haiguangboy.com/posts/transformer-transformer-codesign",
        "relation": "contrast",
        "summary": "Cross-controller reuse remains limited to the same embodiment and requires regenerating supervision, which contradicts the claim that the same DiT switches among generator, designer, and cross-embodiment controller by changing conditional masks",
        "strength": "strong"
      },
      {
        "paper_id": "nvidia_gr00t_n1_7_egoscale_blog_20260417_2026_08",
        "title": "nvidia_gr00t_n1_7_egoscale_blog_20260417",
        "url": "https://haiguangboy.com/posts/nvidia_gr00t_n1_7_egoscale_blog_20260417",
        "relation": "contrast",
        "summary": "The trend of data scaling is insufficient to establish a scaling law, which contradicts NVIDIA GR00T N1.7's bet on the \"scaling of human first-person video\" route: pretrained on 20,000+ hours of EgoScale, officially claimed to have discovered the \"first dexterity scaling law\"—from 1,000 hours to 20,000 hours of data, average task completion more than doubled, without relying on massive teleoperation",
        "strength": "strong"
      },
      {
        "paper_id": "flex_pi_a_multi_stream_world_action_model_with_compute_flexibility_2026_08",
        "title": "Fusing RGB, 3D geometry, and object semantics into a single representation—this is the fundamental reason it is more accurate than π0.5 and Fast-WAM",
        "url": "https://haiguangboy.com/posts/flex_pi_a_multi_stream_world_action_model_with_compute_flexibility",
        "relation": "contrast",
        "summary": "FLEX-π: A More Comprehensive Joint Representation",
        "strength": "strong"
      },
      {
        "paper_id": "dexora_open_source_vla_for_high_dof_bimanual_dexterity_2026_08",
        "title": "Noise in Teleoperation Data Must Enter the Loss Function",
        "url": "https://haiguangboy.com/posts/dexora",
        "relation": "contrast",
        "summary": "Cross-controller reuse remains limited to the same embodiment and requires regenerating supervision, which contradicts the cross-embodiment premise: high-DoF policies treat low-DoF embodiments as subspaces, and dimensionality reduction is easier than dimensionality expansion",
        "strength": "strong"
      }
    ]
  },
  "analyst_take": {
    "type": "author_opinion",
    "text": "💡 Engineering Insights\nSupervision mismatch can also cause negative transfer: after SONIC added converted data without command optimization, the success rate on already-covered conditions was 48.75%, lower than the 62.5% achieved with robot data alone.\nCompared with WholeBodyWAM's cross-controller results, data reusability and policy adaptation-free operation are two separate issues. KODA places reuse at the level of motion and observation records, leaving interface differences to supervision generation. Future work should examine: whether task gains can be preserved and repeated conversion costs reduced under more controllers."
  },
  "ruling": {
    "importance_score": 3.0,
    "one_sentence": "KODA: Converting Human Demonstrations into Humanoid Whole-Body Data"
  },
  "asset_base_url": "https://static.haiguangboy.com/papers/koda",
  "assets": [
    {
      "type": "cover_image",
      "object_key": "papers/koda/cover.webp",
      "url": "https://static.haiguangboy.com/papers/koda/cover.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "post_cover"
    },
    {
      "type": "pdf_screenshot",
      "object_key": "papers/koda/page_01.webp",
      "url": "https://static.haiguangboy.com/papers/koda/page_01.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "paper_first_page"
    },
    {
      "type": "pdf_screenshot",
      "object_key": "papers/koda/key_figure.webp",
      "url": "https://static.haiguangboy.com/papers/koda/key_figure.webp",
      "content_type": "image/webp",
      "upload_status": "pending",
      "role": "method_figure"
    },
    {
      "type": "public_manifest",
      "object_key": "papers/koda/public_manifest.json",
      "url": "https://static.haiguangboy.com/papers/koda/public_manifest.json",
      "content_type": "application/json; charset=utf-8",
      "upload_status": "pending",
      "role": "public_manifest"
    },
    {
      "type": "public_brief",
      "object_key": "papers/koda/public_brief.md",
      "url": "https://static.haiguangboy.com/papers/koda/public_brief.md",
      "content_type": "text/markdown; charset=utf-8",
      "upload_status": "pending",
      "role": "public_brief"
    }
  ],
  "published_at": "2026-10-18T22:26:42+08:00",
  "created_at": "2026-10-18T22:26:42+08:00",
  "updated_at": "2026-10-18T22:26:42+08:00"
}