ImplicitRDP: An End-to-End Visual-Force Diffusion Policy with Structural Slow-Fast Learning

Wendi Chen, Han Xue, Yi Wang, Fangyuan Zhou, Jun Lv, Yang Jin, Shirun Tang, Chuan Wen, Cewu Lu

Keywords: Imitation Learning, Contact-Rich Manipulation, Visual-Force Policy Learning, Slow-Fast Policy Learning

TL;DR: We propose ImplicitRDP, a unified end-to-end visual-force diffusion policy that integrates slow visual planning and fast force control.

Abstract: Contact-rich manipulation requires combining global visual context with local force feedback. We propose ImplicitRDP, an end-to-end visual-force diffusion policy that unifies visual planning and reactive force control in a single network. Our Structural Slow-Fast Learning uses causal attention to process low-frequency visual tokens and high-frequency force tokens for closed-loop control within an action chunk. We further introduce Virtual-target-based Representation Regularization, which predicts a virtual target in the action space to encourage effective use of force feedback and to avoid modality collapse. Experiments on real-world contact-rich tasks show that ImplicitRDP outperforms both vision-only and hierarchical visual-force baselines with a simpler training pipeline.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 2