Mixture of Frames Policy: Multi-Frame Action Denoising for Bimanual Mobile Manipulation

2026-07-13

AI Summary

Existing diffusion-based policies are constrained by a single predefined action coordinate frame, limiting their ability to effectively model the complex action distributions required for bimanual robotic manipulation across diverse tasks. This work proposes a multi-frame synchronous denoising diffusion policy that leverages shared canonical states, dedicated denoising networks, and a cross-frame fusion mechanism to generate higher-quality actions. The method introduces a column-based 6D rotation representation to enable precise and differentiable SE(3) frame transformations under intermediate noisy states, and for the first time achieves simultaneous multi-frame denoising and fusion, breaking away from conventional single-frame modeling paradigms. Evaluated on nine simulated bimanual manipulation tasks and two real-world mobile manipulation tasks, the approach significantly outperforms single-frame baselines, oracle frame selection strategies, and standard mixture-of-experts (MoE) methods.

Abstract

Robotic manipulation is inherently multi-frame: local actions may be simple in an end-effector frame, while transport, upright-object handling, and whole-body coordination are better represented in a base-aligned frame. However, modern diffusion-based visuomotor policies typically commit to a single predefined action frame, forcing one denoiser to model action distributions that are often unnecessarily complex in that frame. We propose Mixture of Frames Policy (MoF), a diffusion policy that performs synchronized action denoising across multiple coordinate frames. MoF maintains a single canonical diffusion state, re-expresses it in several task-relevant frames, applies frame-specialized denoisers, and fuses their noise predictions back in the canonical frame. To make this possible for intermediate noisy diffusion states, we introduce a column-based 6D rotation representation within an SE(3) action parameterization that supports exact, differentiable frame transformations without requiring noisy rotations to lie on the SO(3) manifold. Across nine simulated bimanual manipulation tasks, we show that the best action frame is task-dependent and that MoF improves over oracle frame selection and standard Mixture-of-Experts (MoE) baselines. We further evaluate MoF on two real-world bimanual mobile manipulation tasks, demonstrating that it outperforms all constituent single-frame baselines. Project homepage: https://mofpo.github.io

Problem

• multi-frame action representation
• bimanual manipulation
• diffusion policy
• coordinate frames
• visuomotor policy

Innovation

• Mixture of Frames
• Diffusion Policy
• Multi-Frame Action Denoising
• SE(3) Representation
• Bimanual Manipulation