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

Lifeng Zhuo, Wendi Chen, Han Xue, Shirun Tang, Jun Lv, Cewu Lu, Chuan Wen · N/A · 2026

Summary

Abstract Summary

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.

Key Points

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

Preliminary

We adopt the consistent reactive diffusion inference mechanism from prior end-to-end visual-force diffusion policies [ 4 ] . This mechanism combines slow visual-proprioceptive tokens and fast force tokens in one Transformer. Its structural slow-fast learning separates observations into a slow part for visual context and a fast part for dense force feedback.

Method

FA-RDP addresses the low force-feedback frequency of prior end-to-end policies with a multi-frequency visual-force diffusion backbone. It preserves multimodal pre-contact behavior while enabling high-frequency control through three components: a multi-frequency visual-force Transformer that supports multi-frequency action prediction (Sec.

IV-A ), multimodality-based frequency selection that chooses the appropriate sampling mode (Sec.

Method — FA-RDP: A Frequency-Adaptive Reactive Diffusion Policy for Contact-Rich Manipulation — Fig. 3: Manifold consistency distillation for the high-frequency sampler. A teacher transition maps $\mathbf{A}^{h}_{k}$ to a less noisy state $\mathbf{A}^{h}_{r}$. Both networks predict action chunks on the robot action manifold. MCD aligns the student prediction with the stop-gradient target prediction, while SRL regularizes the student prediction to stay close to the demonstration action.

Experiments

We evaluate FA-RDP on three real-world contact-rich manipulation tasks to answer four key questions: Q1: How does FA-RDP compare against visual-only diffusion policy, hierarchical visual-force policy, fixed-frequency end-to-end visual-force policy, and a non-diffusion regression baseline? Q2: Does FA-RDP preserve multimodal ability?

Q3: Does the multimodality indicator improve performance by switching between frequency modes? Q4: Does our distillation perform better than other distillation methods? Our hardware setup uses a Flexiv Rizon 4R lead...

Experiments — FA-RDP: A Frequency-Adaptive Reactive Diffusion Policy for Contact-Rich Manipulation — Fig. 4: Robot workspace and task objects. The Rizon 4s follower arm executes tasks using observations from a wrist-mounted iPhone camera and a fixed third-view USB camera, while the workspace contains the target box, switch, button, front blocks, and fixture.

Sources and Demonstrations

Figures are reproduced from Zhuo et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.