GraspGen: A Diffusion-based Framework for 6-DOF Grasping with On-Generator Training

TL;DR

GraspGen is a diffusion-based framework that improves 6-DOF robotic grasping by using a novel training method, a large dataset, and a transformer architecture, leading to superior performance in simulation and real-world tests.

Contribution

The paper introduces GraspGen, a diffusion transformer framework with an on-generator discriminator training recipe and a large grasp dataset, advancing 6-DOF grasping capabilities.

Findings

Outperforms prior methods in simulation across various grippers.

Achieves state-of-the-art results on FetchBench benchmark.

Demonstrates effective real robot grasping with noisy visual data.

Abstract

Grasping is a fundamental robot skill, yet despite significant research advancements, learning-based 6-DOF grasping approaches are still not turnkey and struggle to generalize across different embodiments and in-the-wild settings. We build upon the recent success on modeling the object-centric grasp generation process as an iterative diffusion process. Our proposed framework, GraspGen, consists of a DiffusionTransformer architecture that enhances grasp generation, paired with an efficient discriminator to score and filter sampled grasps. We introduce a novel and performant on-generator training recipe for the discriminator. To scale GraspGen to both objects and grippers, we release a new simulated dataset consisting of over 53 million grasps. We demonstrate that GraspGen outperforms prior methods in simulations with singulated objects across different grippers, achieves state-of-the-art performance on the FetchBench grasping benchmark, and performs well on a real robot with noisy visual observations.