code_weights_and_sample_data · 已核验官方资源
审核码：verified_official; 官方关系：verified_apache_code_public_weights_and_mit_sample_with_full_dataset_coming_soon; 最近核验：2026-07-29代码：training, benchmark evaluation and real-robot client/server source
权重：four public safetensors variants for pretrained, LIBERO, LIBERO-plus and RoboCasa
数据：public H-Tac sample only; full 160-hour corpus remains coming soon
许可证：Apache-2.0_code_MIT_sample_model_metadata_missing；Source repository is Apache-2.0; H-Tac sample LICENSE is MIT; Hugging Face model card has no artifact-specific license metadata.
依赖：PyTorch/CUDA、LeRobot-style datasets、LIBERO and RoboCasa for simulation、BeingH/TTP robot client-server stack
硬件/传感器：Wuji dexterous hands、JQ-Industries tactile glove、three RealSense cameras for DeskTask-Tac、CUDA GPU
实机证据：Nine task settings, each with ten in-distribution and five out-of-distribution physical trials.
独立复现：none
限制/反面证据：Full H-Tac dataset coming soon；HF model artifact lacks license metadata；Sample is not paper-scale data；No independent reproduction
当前建议：触觉VLA主流技术2.0·先复算公开权重与样本；Verify one fixed benchmark checkpoint and sample loader before treating the unreleased full H-Tac corpus as available.
重点标签：tactile-vla、code:verified、weights:verified、data:sample-only、real:denominator-backed
核验来源：来源 1 · 来源 2 · 来源 3 · 来源 4 · 来源 5