arXiv:2605.27919v1 Announce Type: cross 
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/

تم تقديم خوارزمية جديدة، وهي مشغل توجيه التردد (FGO)، لتحسين تدريب السياسات البصرية الحركية في المناورة الروبوتية من خلال معالجة الضوضاء عالية التردد الموجودة في العروض البشرية. تتيح هذه الطريقة توليد إجراءات أكثر سلاسة من خلال توجيه عملية الانتشار عبر مناحي تحت التردد، مما يقلل من تضخيم العيوب غير المرغوب فيها.

Se ha introducido un nuevo algoritmo, el Operador de Guía de Frecuencia (FGO), para mejorar el entrenamiento de políticas visuomotoras en la manipulación robótica al abordar el ruido de alta frecuencia presente en las demostraciones humanas. Este método permite una generación de acciones más suave al guiar el proceso de difusión a través de variedades sub-frecuenciales, reduciendo así la amplificación de artefactos indeseables.

Un nouvel algorithme, l'Opérateur de Guidage de Fréquence (FGO), a été introduit pour améliorer l'entraînement des politiques visuomotrices dans la manipulation robotique en s'attaquant au bruit haute fréquence présent dans les démonstrations humaines. Cette méthode permet une génération d'actions plus fluide en guidant le processus de diffusion à travers des variétés sous-fréquentielles, réduisant ainsi l'amplification des artefacts indésirables.

A novel algorithm, the Frequency Guidance Operator (FGO), has been introduced to enhance the training of visuomotor policies in robotic manipulation by addressing the high-frequency noise present in human demonstrations. This method allows for smoother action generation by guiding the diffusion process through sub-frequency manifolds, thereby reducing the amplification of undesirable artifacts.