DextER：具备具身推理的语言驱动机器人灵巧抓取生成
DextER: Language-driven Dexterous Grasp Generation with Embodied Reasoning
把「哪根手指连杆接触物体表面何处」作为具身感知的中间表示，用自回归 token 生成桥接任务语义与物理约束；这一接触 token 化的思路可迁移到其他手-物交互、可供性预测与可控操作生成任务。
原文依据 predicting which hand links contact where on the object surface provides an embodiment-aware intermediate representation
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论语言驱动的灵巧抓取生成要求模型理解任务语义、三维几何以及复杂的手-物交互。尽管视觉语言模型已被应用于该问题，现有方法直接将观测映射为抓取参数，而缺少对物理交互的中间推理。本文提出 DextER（Dexterous Grasp Generation with Embodied Reasoning），为多指操作引入基于接触的具身推理。我们的核心洞见是：预测哪根手部连杆接触物体表面的哪个位置，可提供一种具身感知的中间表示，从而在任务语义与物理约束之间建立桥梁。DextER 以自回归方式先生成具身接触 token，指明各手指连杆在物体表面上的接触位置，随后生成编码手部构型的抓取 token。在 DexGYS 上，DextER 取得 67.14% 的成功率，以 3.83 个百分点的优势超过最优方法，意图对齐提升 96.4%。我们还通过部分接触指定展示了可引导生成，从而对抓取合成提供细粒度控制。
英文原文摘要 1137 字符
Language-driven dexterous grasp generation requires the models to understand task semantics, 3D geometry, and complex hand-object interactions. While vision-language models have been applied to this problem, existing approaches directly map observations to grasp parameters without intermediate reasoning about physical interactions. We present DextER, Dexterous Grasp Generation with Embodied Reasoning, which introduces contact-based embodied reasoning for multi-finger manipulation. Our key insight is that predicting which hand links contact where on the object surface provides an embodiment-aware intermediate representation, bridging task semantics with physical constraints. DextER autoregressively generates embodied contact tokens specifying which finger links contact where on the object surface, followed by grasp tokens encoding the hand configuration. On DexGYS, DextER achieves 67.14% success rate, outperforming state-of-the-art by 3.83 p.p. with 96.4% improvement in intention alignment. We also demonstrate steerable generation through partial contact specification, providing fine-grained control over grasp synthesis.