DexVLG: Dexterous Vision-Language-Grasp Model at Scale

TL;DR

DexVLG is a large-scale vision-language model designed for dexterous robotic grasping, trained on an extensive dataset, demonstrating strong zero-shot generalization and real-world applicability for human-like hand manipulation tasks.

Contribution

The paper introduces DexVLG, a novel large-scale vision-language model for dexterous grasping, supported by a new extensive dataset DexGraspNet 3.0 and benchmarks for evaluation.

Findings

Achieves over 76% zero-shot success rate in simulation

Sets new state-of-the-art part-grasp accuracy in simulation

Demonstrates successful real-world part-aligned grasping

Abstract

As large models gain traction, vision-language-action (VLA) systems are enabling robots to tackle increasingly complex tasks. However, limited by the difficulty of data collection, progress has mainly focused on controlling simple gripper end-effectors. There is little research on functional grasping with large models for human-like dexterous hands. In this paper, we introduce DexVLG, a large Vision-Language-Grasp model for Dexterous grasp pose prediction aligned with language instructions using single-view RGBD input. To accomplish this, we generate a dataset of 170 million dexterous grasp poses mapped to semantic parts across 174,000 objects in simulation, paired with detailed part-level captions. This large-scale dataset, named DexGraspNet 3.0, is used to train a VLM and flow-matching-based pose head capable of producing instruction-aligned grasp poses for tabletop objects. To assess DexVLG's performance, we create benchmarks in physics-based simulations and conduct real-world experiments. Extensive testing demonstrates DexVLG's strong zero-shot generalization capabilities-achieving over 76% zero-shot execution success rate and state-of-the-art part-grasp accuracy in simulation-and successful part-aligned grasps on physical objects in real-world scenarios.