code_data_and_deployment_state · 已核验官方资源
审核码：verified_official; 官方关系：verified_official_partial_release; 最近核验：2026-07-23代码：Fixed commit f92a6925fd8a87efff3bd6201cd5f1d47d2f2d2f exposes Isaac Lab 2.2 collection, PICO teleoperation, record/replay and randomized path-tracing code.
权重：A 4.2 MB TeleopIt tracking ONNX is bundled; the paper's high-level flow-matching planner checkpoint and real-robot deployment stack are absent.
数据：Object and texture downloads are linked, but the complete four-task 200-trajectory corpus and 4,000 randomized renderings are not released.
许可证：MIT_with_external_asset_terms；MIT covers repository-owned source; PICO, XRoboToolkit, GMR, Hunyuan3D, Isaac and cloud-hosted object/texture assets retain upstream terms.
依赖：Ubuntu 22.04、Isaac Lab 2.2、PICO SDK and XRoboToolkit、GMR、external object and texture archives
硬件/传感器：RTX-class GPU for path tracing、PICO headset and controllers、Paper deployment uses Unitree G1, dexterous hands, one D435i, two D405 cameras and RTX 4090
实机证据：Four G1 tasks, ten trials each, report final zero-shot results of 8/10, 7/10, 8/10 and 10/10.；Simulation collection takes 15.2–28.4 minutes for 50 successes per task versus 17.5–44.8 minutes on the real robot.
独立复现：No independent OASIS policy or G1 reproduction was found. The current deployment-code issue is an unanswered release request, not reproduction evidence.
限制/反面证据：The complete training corpus, high-level policy and physical deployment code are absent.；Only successful demonstrations are counted; failure trajectories and contact-force labels are not released.；Visual randomization leaves motion trajectories unchanged and evaluates only four author tasks.
当前建议：高潜待持续验证；The public collection/replay stack is useful for a bounded internal data-pipeline trial, but it cannot reproduce the paper policy or physical deployment.
重点标签：仿真遥操作与视觉随机化主流迭代、artifact:partial-release、dataset:not-released、deployment:not-released、community:unverified
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4