Search papers, labs, and topics across Lattice.
This paper introduces FA-RDP, a frequency-adaptive reactive diffusion policy designed to optimize contact-rich manipulation by balancing action multimodality and reactivity. By employing a shared multi-frequency visual-force Transformer, FA-RDP dynamically selects between low-frequency, multi-step sampling before contact and high-frequency, one-step sampling as ambiguity decreases, effectively addressing the tradeoff inherent in standard diffusion policies. Experimental results demonstrate that FA-RDP not only achieves the highest success rates across various tasks but also maintains diverse pre-contact trajectory modes, highlighting its effectiveness in complex manipulation scenarios.
FA-RDP achieves superior success rates in contact-rich manipulation while preserving diverse action modes, revolutionizing how we approach multimodal decision-making in robotics.
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.