code_weights_and_linked_latent_data · 已核验官方资源
审核码：verified_official; 官方关系：verified_official_calvin_code_and_author_weights_with_incomplete_real_robot_path; 最近核验：2026-07-23代码：Fixed commit 43e4dae8e9461f5d6b7e3760f1b728e083ff86ed contains 162 tracked files, including CALVIN inference/evaluation, GRPO post-training and action-policy training.
权重：VAMPO_SVD revision dc862ed62fa94720030a6840d5258f03ea8cfd8b and VAMPO_policy revision 5a15d269e73ab160d499b1dff77d7d01605dcc6f are public and non-gated.
数据：The README-linked vpp_svd_latent revision a9f9d7faff502da3875daa9e7ee137fe2cf91e4c is public, but has no declared dataset license. VAMPO's 600 AgiBot demonstrations are not released.
许可证：MIT_code_Apache-2.0_author_weights_dataset_terms_missing；MIT covers repository code. The two author model cards declare Apache-2.0. The upstream baseline weight and vpp_svd_latent dataset do not declare licenses, and CALVIN retains separate upstream terms.
依赖：Python 3.10、PyTorch 2.1.1、diffusers 0.28.0、transformers 4.47.0、Accelerate、CALVIN environment and dataset、Hydra、decord
硬件/传感器：64 NVIDIA H20 GPUs for reported VPM post-training、8 NVIDIA H20 GPUs for reported AGM training、NVIDIA RTX 5880 evaluation、AgiBot Genie 01 with D455 and two D405 cameras for matching physical trials
实机证据：The paper reports three AgiBot tasks and 200 teleoperated demonstrations per task, but no repeated-test denominator or confidence interval.；step3_deploy_real_xbot.py loads a policy and processes three image streams, but retains author-local absolute paths and does not provide a complete robot transport or safety layer.
独立复现：No independent fixed-commit CALVIN reproduction, H20 training report or AgiBot deployment was found.
限制/反面证据：No repository tests or pinned environment lock beyond requirements.txt.；The real-robot script is not a complete deployable AgiBot integration.；Reported VPM training requires 64 H20 GPUs.；AgiBot demonstrations and real evaluation logs are absent.；The linked latent dataset and baseline checkpoint lack declared licenses.；No independent reproduction.
当前建议：近期热门·高成本可复查；Run a bounded CALVIN evaluation from the fixed revisions before considering any physical adapter. Do not present the author-local Xbot script as turnkey AgiBot deployment.
重点标签：视频动作模型主流技术2.0、可能成为新主流、artifact:code+weights、deployment:calvin-first、license:layered、community:unverified
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4 · 来源 5 · 来源 6
