Event

OpenDexGrasp proposes an open-vocabulary task-oriented dexterous grasping framework

1 reports · 1 sources

Jiyao Zhang et al. propose OpenDexGrasp, a unified data and generative modeling framework for open-vocabulary task-oriented dexterous grasp generation, enabling robots to infer functional intent from free-form language and, combining multi-view visual observations with object geometry, generate executable high-DOF grasps. The paper was accepted by CoRL 2026. The accompanying data framework OpenDexVerse provides dual-source supervision: OpenDex-Scale provides large-scale semantic and geometric coverage through automated grasp synthesis and vision-language annotation, while OpenDex-Align provides high-quality embodied alignment through human teleoperation and category-level transfer. The authors report that simulation and real-robot experiments show improvements in functional alignment, physical feasibility, generalization to unseen categories, and real-world execution success rate.