mixed · verified_hold
审核码:verified_hold; 官方关系:paper_only_real_robot_and_dataset_claim_no_method_artifact; 最近核验:2026-08-15代码:not released in the paper's official arXiv record; no official source repository was linked or independently located in the 2026-08-15 primary-source review
权重:not released
数据:XHT-Dataset is described as 900 real-world demonstrations across seven contact-rich tasks on XHand--UR7e, but no download endpoint, data card, access procedure, format, license, calibration package, or success/failure split was published
许可证:not_disclosed_for_method_artifacts;The arXiv article distribution permission is not a license for absent source code, weights, XHT-Dataset, robot drivers, calibration files, or tactile hardware designs.
依赖:VLA policy with tactile-patch encoder、online tactile prediction/refinement and high-frequency action module、unreleased XHT-Dataset and hardware calibration
硬件/传感器:XHand five-finger dexterous hand、Universal Robots UR7e arm、tactile sensing used in the closed loop
实机证据:The official abstract reports closed-loop real-robot experiments on the XHand--UR7e platform.;It reports average success-rate gains of 18.4 percentage points under standard conditions and 23.8 percentage points under challenging conditions versus its strongest baseline; trial denominators, per-task counts, raw logs and failure taxonomy are not available from the official record.
独立复现:None found in the 2026-08-15 review. Author-reported hardware experiments are not counted as community reproduction.
限制/反面证据:The claimed XHT-Dataset is not downloadable or licensable from a primary endpoint.;No method code, checkpoint, environment lock, control frequency, tactile calibration or safety layer is inspectable.;Reported aggregate gains cannot be compared fairly with other tactile systems without per-task denominator, object split and challenge protocol.
当前建议:真机证据待开放;This is a relevant real dexterous-hardware signal with an explicitly described data collection scale, but it is not a deployable or reproducible framework until official code, XHT-Dataset access and hardware timing/calibration evidence are released.
重点标签:视觉-触觉VLA主流技术迭代、真实五指灵巧手、真实机械臂、在线触觉预测、近期热门、待社区验证
核验来源:来源 1