FA-RDP: A Frequency-Adaptive Reactive Diffusion Policy for Contact-Rich Manipulation

2026-07-30

AI Summary

This work addresses the challenge that standard diffusion policies face in contact-rich manipulation tasks, where a fixed inference frequency struggles to simultaneously capture the multimodality of pre-contact motions and enable rapid force-responsive reactions post-contact. To overcome this limitation, the authors propose a frequency-adaptive reactive diffusion policy that dynamically switches sampling frequencies via a learnable multimodal indicator: multi-step low-frequency sampling preserves action diversity before contact, while single-step high-frequency sampling enhances responsiveness after contact. A multi-frequency visuo-tactile Transformer is introduced to unify perceptual inputs across frequencies, and manifold consistency distillation (MCD) is employed to constrain action predictions to the robot’s motion manifold while retaining DDPM residual supervision. The method achieves state-of-the-art success rates across three contact-intensive tasks and effectively maintains diverse pre-contact trajectory modes.

Abstract

In contact-rich manipulation, action multimodality and reactivity dominate different stages of a single episode. Before contact, multiple trajectories might be equally valid, making it important to preserve diverse action modes. After contact, geometric constraints and force limits narrow the solution space, while successful execution demands rapid responses to force feedback. However, standard diffusion policies use a fixed inference frequency and sampling steps throughout the episode, forcing a fundamental compromise: low-frequency, multi-step sampling better preserves pre-contact multimodality but responds slowly to force feedback, whereas high-frequency sampling improves reactivity but tends to collapse distinct pre-contact modes. To resolve this tradeoff, we present FA-RDP, a frequency-adaptive reactive diffusion policy. A shared multi-frequency visual-force Transformer predicts action chunks at both low and high frequencies, while a learned multimodality indicator dynamically selects multi-step low-frequency sampling before contact and one-step high-frequency sampling as action ambiguity decreases. We further introduce Manifold Consistency Distillation (MCD), which reparameterizes the diffusion network to predict actions on the robot action manifold while retaining DDPM-based residual supervision. Experiments on three contact-rich manipulation tasks show that FA-RDP achieves the highest success rate while preserving diverse pre-contact trajectory modes. Code and videos are available at https://fa-rdp.github.io.

Problem

• contact-rich manipulation
• action multimodality
• reactivity
• diffusion policy
• inference frequency

Innovation

• frequency-adaptive diffusion
• reactive manipulation
• multimodal action generation
• manifold consistency distillation
• visual-force transformer