DiffusionRL: Efficient Training of Diffusion Policies for Robotic Grasping Using RL-Adapted Large-Scale Datasets

TL;DR

This paper introduces DiffusionRL, a method that combines diffusion models and reinforcement learning to efficiently train robotic grasping policies using large datasets, achieving high success rates without manual data collection.

Contribution

The paper presents an end-to-end pipeline that enhances large datasets with RL, enabling effective diffusion policy training for robotic grasping tasks.

Findings

Achieved 80% success rate on three DexGraspNet objects.

Eliminated manual data collection, reducing barriers to diffusion model adoption.

Improved generalization and robustness in robotic manipulation.

Abstract

Diffusion models have been successfully applied in areas such as image, video, and audio generation. Recent works show their promise for sequential decision-making and dexterous manipulation, leveraging their ability to model complex action distributions. However, challenges persist due to the data limitations and scenario-specific adaptation needs. In this paper, we address these challenges by proposing an optimized approach to training diffusion policies using large, pre-built datasets that are enhanced using Reinforcement Learning (RL). Our end-to-end pipeline leverages RL-based enhancement of the DexGraspNet dataset, lightweight diffusion policy training on a dexterous manipulation task for a five-fingered robotic hand, and a pose sampling algorithm for validation. The pipeline achieved a high success rate of 80% for three DexGraspNet objects. By eliminating manual data collection, our approach lowers barriers to adopting diffusion models in robotics, enhancing generalization and robustness for real-world applications.