partial_code · 暂缓：缺少资源级许可证
审核码：hold_no_artifact_scoped_license; 官方关系：verified_official_partial_release_without_license; 最近核验：2026-07-20代码：Training source and G1 LiDAR task configuration are public on top of Active Adaptation.
权重：No policy checkpoint was verified.
数据：No training dataset is expected for the simulator-generated LiDAR path, but assets and evaluation outputs are not packaged as a release.
许可证：not_disclosed；No LICENSE file or package license declaration was found. The upstream Active Adaptation license cannot be inherited automatically.
依赖：Active Adaptation framework、distributed PPO training、Gallant Python package
硬件/传感器：README example uses 8 GPUs and 1024 environments、Unitree G1 with LiDAR for the reported physical path
实机证据：Official paper/project shows physical humanoid navigation through stairs, narrow passages, overhead constraints and multi-level terrain.
独立复现：No independent G1+LiDAR reproduction with commit and comparable success metrics was found.
限制/反面证据：README explicitly says detailed play/evaluation commands will be released later.；No license, checkpoint, packaged environment or independent reproduction exists.；The documented 8-GPU example is not a low-cost training route.
当前建议：高潜待持续验证；High technical relevance for navigation, but keep out of reproducible rankings until evaluation commands, license and checkpoint are available.
重点标签：recency:2025、navigation:lidar-voxel、locomotion:3d-constrained-terrain、artifact:training-only-partial、cost:8gpu-example、license:missing
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4