Frequency-Guided Action Diffusion via Sub-Frequency Manifold Traversal

2026-05-26

AI Summary

This work addresses the challenge that human demonstrations in behavioral cloning often contain high-frequency noise—such as jitter and hesitations—which can lead imitation policies to learn suboptimal actions, with diffusion-based policies particularly prone to amplifying such artifacts. To mitigate this, the authors propose the Frequency-Guided Operator (FGO), which introduces sub-frequency manifold traversal into diffusion policies for the first time. FGO implicitly shapes the spectral distribution during denoising via a progressive spectral band expansion mechanism, effectively removing high-frequency noise while preserving task-critical fine-grained motion details. Experiments across 15 robotic manipulation tasks spanning five benchmarks demonstrate that FGO significantly enhances action smoothness and temporal consistency without compromising task success rates.

Abstract

Learning visuomotor policies via behavior cloning typically involves mimicking expert demonstrations collected by human operators. However, natural human demonstrations inherently contain high-frequency noise, such as intermittent jerks, pauses, and action jitter. Training policies to directly imitate these raw trajectories inevitably causes the model to inherit these suboptimal behaviors. This pathology is particularly pronounced in diffusion-based policies, where iterative denoising steps can inadvertently amplify high-frequency artifacts at the expense of meaningful fine-grained details. To address these limitations, we present a novel frequency-based algorithm that enables implicit spectral maneuvering and smooth action generation. Our method, Frequency Guidance Operator (FGO), steers the generation process of diffusion polices by progressively driving the noisy samples through intermediate sub-frequency manifolds with expanding spectral bands. Validated on 15 robotic manipulation tasks from 5 benchmarks, FGO achieves superior performance in enhancing action smoothness and temporal consistency while preserving the details necessary for successful task execution. Project website: https://henrywjl.github.io/frequency-guidance-operator/

Problem

• behavior cloning
• high-frequency noise
• diffusion policies
• action smoothness
• temporal consistency

Innovation

• frequency-guided diffusion
• sub-frequency manifold
• visuomotor policy
• behavior cloning
• action smoothness