MATE: Asynchronous Multi-Rate Diffusion Policy Composition

This paper introduces LAG-Fusion, a framework for asynchronous multimodal diffusion policy composition that allows modality-specific policies to operate at their native inference rates and aligns delayed guidance through reference-frame rebasing. The authors instantiate the method by composing a low-frequency vision policy with a high-frequency force policy for contact-rich manipulation, showing improved responsiveness and task performance over synchronous fusion and force-aware baselines under heterogeneous modality latencies.