code_weights_and_data · 已核验官方资源
审核码：verified_official; 官方关系：verified_mit_source_with_reachable_policy_scene_and_dataset_archives; 最近核验：2026-07-28代码：42 tracked entries plus pinned RoboCasa and robomimic submodules
权重：15 manipulation-policy archives plus BC-with-navigation and 3D Gaussian Splatting downloads
数据：five task-specific policy-training archives are referenced by the official downloader
许可证：MIT；Applies to the Mobi-π code; RoboCasa, robomimic, MimicGen, checkpoints and downloaded data retain their own or undisclosed terms.
依赖：Python 3.10、PyTorch 2.1.1、NumPy 1.23.5、timm 1.0.12、Nerfstudio/gsplat、RoboCasa、robomimic、MimicGen
硬件/传感器：CUDA GPU for 3DGS and policy training、mobile dual-arm robot with Basler RGB camera for paper physical setup
实机证据：Three physical grocery tasks, ten initializations per method; Mobi-π full success is 7/10, 10/10 and 8/10.
独立复现：setup failures only
限制/反面证据：Static-scene assumption；Needs manipulation-policy training observations；Dependency conflicts；Checkpoint server previously failed for users；No independent physical reproduction
当前建议：策略感知移动操作主流迭代·先做下载与环境烟测；Pin the release commit, checksum one policy archive and one 3DGS archive, then reproduce one simulation task before using physical hardware.
重点标签：mobile-manipulation、code:verified、weights:verified、data:verified、community:setup-failure
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4 · 来源 5