发表机构
Shanghai Jiao Tong University; Shanghai Innovation Institute; Noematrix Ltd.（上海交通大学; 上海创新研究院; 诺玛特里斯有限公司）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
FA-RDP 是一种频率自适应反应式扩散策略，通过多频率视觉-力 Transformer 和多模态指标动态调整采样策略，结合流形一致性蒸馏，在接触丰富操作任务中兼顾多模态保留与反应性，实现最高成功率。
AI 中文摘要
在接触丰富的操作中，动作多模态与反应性在单个 episode 的不同阶段占据主导地位。接触前，多条轨迹可能同样有效，因此保留多样化的动作模式十分重要；接触后，几何约束和力限会缩小解空间，成功执行则需要对力反馈做出快速响应。然而，标准扩散策略在整个 episode 中使用固定的推理频率和采样步数，这导致了根本性的权衡：低频多步采样能更好地保留接触前的多模态，但对力反馈的响应缓慢；而高频采样虽能提高反应性，却容易使接触前的不同模式坍缩。为解决这一权衡问题，我们提出了 FA-RDP，即频率自适应反应式扩散策略。一个共享的多频率视觉-力 Transformer 可在低频和高频下预测动作块，同时，一个学习得到的多模态指标会动态选择接触前的多步低频采样，并在动作歧义降低时选择单步高频采样。我们进一步引入了流形一致性蒸馏（MCD），该方法对扩散网络进行重参数化，使其在基于 DDPM 的残差监督下，预测机器人动作流形上的动作。在三个接触丰富的操作任务上进行的实验表明，FA-RDP 在保留多样化接触前轨迹模式的同时，达到了最高的成功率。代码和视频可在此 https URL 获取。
英文摘要
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.
CommentsProject page: https://fa-rdp.github.io
