- repository_placeholder · 暂缓：仅占位，尚未发布
审核码：hold_placeholder_not_release; 官方关系：verified_official_placeholder; 最近核验：2026-07-20代码：Not released. The repository contains one README file only.
权重：Not released.
数据：Behavioral data are described by the paper/project but no archive was verified.
许可证：not_disclosed；No LICENSE file or artifact-specific terms exist in the placeholder repository.
依赖：未披露
硬件/传感器：Official paper reports simulation and real-humanoid deployment, but deployable hardware requirements are not released
实机证据：Official paper/project reports real-humanoid control and MPKPE improvements; no released checkpoint or executable path was available for verification.
独立复现：None found; the artifact did not yet exist long enough to support an independent reproduction.
限制/反面证据：A release promise is not an open-source artifact.；No code, weights, dataset, license, environment lock or reproducible command is available.
当前建议：高潜待持续验证；Track daily through the announced 2026-07-26 release window; do not place in a practical ranking until executable code, license and at least one checkpoint are actually present.
重点标签：recency:critical、whole-body:foundation-model、artifact:promised-not-released、evidence:official-real-robot、practice:not-yet-runnable
核验来源：来源 1 · 来源 2 · 来源 3
- code_partial_release · 已核验官方资源
审核码：verified_official; 官方关系：verified_official_partial_source_without_root_license; 最近核验：2026-07-21代码：ScaleRetarget now contains Python source, dataset loaders, G1 correspondence/configuration, format converters, visualization and 12 documented motion-source recipes.
权重：No ScaleBFM policy checkpoint or deployable controller weight was verified.
数据：The repository documents upstream datasets and conversion recipes; it does not release the reported 102-million-frame unified ScaleBFM corpus as one licensed archive.
许可证：not_declared；No root LICENSE file was present at commit b40819fec30f96da2c8b80acd70db994d0540b27. Unitree G1 assets and a vendored LAFAN utility include component-specific notices that do not license the full ScaleRetarget codebase.
依赖：Python 3.11、PyTorch、MuJoCo、Hydra、SMPL-X body models for several source datasets、dataset-specific source downloads and licenses
硬件/传感器：CPU or GPU workstation for offline motion retargeting、Unitree G1 model/configuration is included for visualization and trajectory export、No physical-robot controller, safety layer or deployment package is included
实机证据：The paper and project page report physical humanoid tracking, goal reaching and disturbance recovery; the newly released ScaleRetarget module itself is an offline data-preparation tool and does not reproduce those physical experiments.
独立复现：No independent team report with a fixed commit, dataset license, retargeting metrics and physical G1 deployment was found.
限制/反面证据：The root repository has no artifact-scoped license, so source availability is not equivalent to open-source reuse permission.；ScaleRetarget is only the motion-retargeting component; training, policy weights and physical deployment remain unreleased.；Several source datasets and SMPL-X assets require separate access and terms.；The README explicitly warns about jitter, floating, stiffness and incomplete preparation steps.
当前建议：高潜待持续验证；The retargeting component is now testable and materially improves the prior placeholder state, but the project cannot be treated as fully open or end-to-end reproducible until a root license, training code, policy checkpoints and physical deployment path are released.
重点标签：recency:critical、whole-body:foundation-model、artifact:partial-code-release、retargeting:g1、license:not-declared、weights-data:not-released
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4
- code_partial_release · 已核验官方资源
审核码：verified_official; 官方关系：verified_official_partial_source_without_root_license; 最近核验：2026-07-21代码：ScaleRetarget loaders, retargeting, formatting and visualization are present at the corrected official commit.
权重：No ScaleBFM policy checkpoint, TensorRT engine or deployable controller weight was verified.
数据：The repository documents upstream motion-source conversion but does not release the unified 102-million-frame corpus, Xsens test set or paper split manifest.
许可证：fragmented_no_root_license；No repository-root LICENSE was present. ScaleRetarget package metadata declares MIT and bundled Unitree assets declare BSD-3-Clause, but component notices do not license the full repository or external SMPL-X/motion datasets.
依赖：Python 3.11、PyTorch、MuJoCo、Mink、Hydra、SMPL-X and source-dataset-specific assets
硬件/传感器：offline retargeting workstation、Unitree G1 29-DoF target configuration、no physical controller, safety layer or deployment package in the released component
实机证据：The paper and project page show Unitree G1 deployment at 50 Hz high-level and 200 Hz low-level control, but no task-level physical trial counts or failure statistics are published.
独立复现：No independent fixed-commit ScaleBFM training or physical G1 reproduction with comparable metrics was verified.
限制/反面证据：The corrected commit proves only the partial ScaleRetarget release, not end-to-end reproduction of the paper controller.；The root repository has no unified license and external motion/SMPL-X terms remain separate.；Training, policy weights, unified data, TensorRT export and SDK2 deployment are still unreleased.
当前建议：高潜待持续验证；Use the corrected commit for offline retargeting experiments; do not claim ScaleBFM training, policy or physical deployment reproducibility until those artifacts ship.
重点标签：主流技术2.0、artifact:partial-code-release、commit:corrected、license:fragmented、weights-data:not-released
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4