Grasp as You Say

iSEE Laboratory / DexGYS team

Training, inference and visualization for generating dexterous grasps conditioned on language.

What you can explore

Explore how a text instruction changes the intended grasp, alongside the separate DexGYS dataset record.

Before you use it

Setup requires DexGYS labels, a Shadow Hand model and OakInk object meshes. Language-conditioned generation does not prove task completion on a real robot.

Original contributors

Yi-Lin Wei, Jian-Jian Jiang, Chengyi Xing, Xiantuo Tan, Xiao-Ming Wu, Hao Li, Mark Cutkosky and Wei-Shi Zheng

License / access: Code: MIT · check DexGYS, OakInk and model licenses separately

Original code & documentation

Research overview

Editorial reference · checked 2026-09-22