From Noise to Intent: Anchoring Generative VLA Policies with Residual Bridges

TL;DR

ResVLA introduces a novel refinement-based generative approach that decouples global intent from local dynamics, improving efficiency, robustness, and convergence in embodied robotic control tasks.

Contribution

It proposes ResVLA, a spectral analysis-based architecture that shifts from noise generation to intent refinement, enhancing generative policy performance and robustness.

Findings

ResVLA achieves competitive performance in simulation.

It demonstrates robustness to language and embodiment perturbations.

ResVLA converges faster than standard generative models.

Abstract

Bridging high-level semantic understanding with low-level physical control remains a persistent challenge in embodied intelligence, stemming from the fundamental spatiotemporal scale mismatch between cognition and action. Existing generative VLA policies typically adopt a "Generation-from-Noise" paradigm, which disregards this disparity, leading to representation inefficiency and weak condition alignment during optimization. In this work, we propose ResVLA, an architecture that shifts the paradigm to "Refinement-from-Intent." Recognizing that robotic motion naturally decomposes into global intent and local dynamics, ResVLA utilizes spectral analysis to decouple control into a deterministic low-frequency anchor and a stochastic high-frequency residual. By anchoring the generative process on the predicted intent, our model focuses strictly on refining local dynamics via a residual diffusion bridge. Extensive simulation experiments show that ResVLA achieves competitive performance, strong robustness to language and robot embodiment perturbations, and faster convergence than standard generative baselines. It also demonstrates strong performance in real-world robot experiments.