外观
论文HF Daily Papers · arXiv 2610.2545910 天前
BiDex-100发布：含8080条遥操作数据的双手灵巧操作仿真基准
本文由 AI 根据公开内容整理，可能有误，请以原文为准：BiDex-100: A Comprehensive Benchmark for Bimanual Dexterous Manipulation Across 100+ Simulation Tasks（HF Daily Papers · arXiv 2610.25459）
核心内容
为解决现有机器人动作模型在双手灵巧操作能力表征不足的问题，研究团队推出BiDex-100仿真基准。该基准包含106项任务，沿泛化性、指令遵循、少样本学习、长时序规划及记忆能力五个维度构建评估体系。数据集提供8080条经成功验证的遥操作演示数据，总计625万帧，支持两种高自由度手部本体，并配套域外变体测试以量化模型鲁棒性。
技术层面，BiDex-100通过受控的视觉与空间变化设计域外测试集，系统评估模型在非结构化环境下的适应能力。研究基于统一协议对7款最新Vision-Language-Action模型进行评测，确立了涵盖特定能力性能、鲁棒性及数据效率的基准线。该项目已开源网站与数据资源，为具身智能算法训练提供了标准化的双手灵巧操作验证环境与高质量真值数据支撑。
关键看点
- 任务体系：涵盖106项仿真任务，沿泛化、指令遵循、少样本、长时序及记忆五轴评估双手灵巧操作能力
- 数据规模：提供8080条成功验证的遥操作演示，累计625万帧数据，支持两种高自由度手部本体
- 鲁棒性测试：每项任务配对域外变体，通过受控视觉与空间变化系统评估模型在非分布环境下的适应性
- 模型基线：基于统一协议评测7款主流VLA模型，确立能力特异性性能、鲁棒性及数据效率基准
工程与产业观察
BiDex-100填补了双手灵巧操作标准化评测空白，其625万帧遥操作数据与五维能力评估体系为VLA模型训练提供了关键Ground Truth。对于具身智能开发者，该基准不仅验证了当前模型在复杂操作中的短板，更为解决Sim-to-Real迁移中的数据稀缺与泛化难题提供了可复现的工程参照系。
英文摘要 (Abstract)
Bimanual manipulation with dexterous hands has attracted growing attention in research on robotic action models. However, few studies have developed benchmarks specifically tailored for dexterous manipulation, leaving the dexterous manipulation capabilities of existing models insufficiently characterized. In this paper, we present BiDex-100, a simulation benchmark for bimanual dexterous manipulation that combines broad task coverage, multi-dimensional evaluation, and high-DoF hand embodiments. Concretely, the benchmark comprises 106 tasks organized along five capability axes: generalization, instruction following, limited demonstrations, long horizon, and memory. Each task is paired with an out-of-domain variant featuring controlled visual and spatial changes, allowing systematic assessment of robustness. In addition, the benchmark supports two embodiments and provides 8,080 success-validated teleoperated demonstrations, totaling 6.25 million frames. We evaluate seven recent Vision-Language-Action models under a unified protocol, establishing baselines for capability-specific performance, robustness, and data efficiency. The website is available at https://BiDex-100.github.io/BiDex-100/.
Hugging Face Daily Papers 热度：2 upvotes 作者：Yong Wang, Kecheng Zheng, Xing Zhu, Chao Xu, Yujun Shen, Shengze Cai
来源
本页事实对应的公开页面，以原文为准。