code_and_data · 已核验官方资源
审核码：verified_official; 官方关系：gated_noncommercial_dataset_and_unlicensed_training_code; 最近核验：2026-07-29代码：training and data-processing source is public in a one-commit repository
权重：paper checkpoints not released
数据：2.36 TB gated dataset; 50,129-episode paper snapshot
许可证：custom_noncommercial_data_and_unlicensed_code；Dataset is non-commercial, no redistribution and revocable access; GitHub training repository has no LICENSE.
依赖：OpenPI at c23745b5ad24e98f66967ea795a07b2588ed6c79、LeRobot v3 format、conda environment、AXIS dataset access
硬件/传感器：paper pretraining uses 8 A100 GPUs、browser collection uses commodity client hardware
实机证据：未披露
独立复现：none
限制/反面证据：Simulated Franka focus；No quantified sim-to-real；Gated non-commercial dataset；No code license；No released paper checkpoints
当前建议：具身数据引擎主流技术2.0·受限开放；Request a small snapshot, audit exact terms and replay success, then run a one-task OpenPI smoke test.
重点标签：robot-data-engine、data:gated、code:license-missing、real:none
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4
