UniVTAC Unified Visuo-Tactile Simulation Platform and Benchmark
UniVTAC 是基于 NVIDIA Isaac Lab 与 TacEx/UIPC 触觉模拟的视觉-触觉数据生成、策略训练与评测平台。官方项目页列出 GelSight Mini、ViTai GF225、Xense WS 三类传感器和八项接触操作任务；官方仓库提供安装、数据采集、ACT/ViTaL 基线与统一评测入口，数据仓提供每任务100 episodes 的下载入口。
- 开放状态
- 项目制品开放·上游条款另适用
- 许可证
- MIT for the official repository; dataset distribution and upstream sensor/simulator assets require separate terms review
- 商业使用
- repository MIT does not automatically relicense TacEx, Isaac Lab, sensor assets, downloaded episodes or upstream dependencies
- 人工评审结论
- official_open_MIT_platform_with_public_episode_routes_and_dataset_schema_failure
- 机器人
- simulated manipulation arm、Tianji Marvin real-world validation platform
- 任务
- Lift Bottle, Lift Can, Put Bottle in Shelf, Grasp Classify, Insert Hole, Insert Tube, Insert HDMI, Pull Out Key
- 模态
- RGB、depth、simulated tactile RGB/depth/marker/pose、proprioception、actions、contact/force-related signals
- 格式
- repository task files plus Hugging Face/ModelScope dataset distributions; HF viewer currently reports DatasetGenerationError/CastError schema mismatch
- 采集/生成方式
- GPU-accelerated Isaac Lab/TacEx simulation with scripted high-level manipulation APIs and expert demonstration collection
- 已知问题
- HF dataset viewer currently fails schema casting; complete download, file hashes, dataset card revision, synchronization, calibration, failure data, privacy and independent sim-to-real reproduction remain unverified. Do not treat the MIT code license as a blanket license for all assets.
论文原文 · 官方项目页 · 官方代码固定入口 · 官方 Hugging Face 数据入口 · 官方 ModelScope 数据入口