Two-Steps Diffusion Policy for Robotic Manipulation via Genetic Denoising

TL;DR

This paper introduces a tailored diffusion policy for robotic manipulation that uses a novel genetic denoising strategy, enabling high performance with minimal neural function evaluations and outperforming standard methods across multiple tasks.

Contribution

The work adapts diffusion models for embodied AI by designing a genetic denoising approach that improves efficiency and stability in robotic control tasks.

Findings

Achieves up to 20% performance gains over standard diffusion policies.

Operates effectively with as few as 2 neural function evaluations.

Consistently outperforms existing methods across 14 manipulation tasks.

Abstract

Diffusion models, such as diffusion policy, have achieved state-of-the-art results in robotic manipulation by imitating expert demonstrations. While diffusion models were originally developed for vision tasks like image and video generation, many of their inference strategies have been directly transferred to control domains without adaptation. In this work, we show that by tailoring the denoising process to the specific characteristics of embodied AI tasks -- particularly structured, low-dimensional nature of action distributions -- diffusion policies can operate effectively with as few as 5 neural function evaluations (NFE). Building on this insight, we propose a population-based sampling strategy, genetic denoising, which enhances both performance and stability by selecting denoising trajectories with low out-of-distribution risk. Our method solves challenging tasks with only 2 NFE while improving or matching performance. We evaluate our approach across 14 robotic manipulation tasks from D4RL and Robomimic, spanning multiple action horizons and inference budgets. In over 2 million evaluations, our method consistently outperforms standard diffusion-based policies, achieving up to 20\% performance gains with significantly fewer inference steps.