CF-VLA: Efficient Coarse-to-Fine Action Generation for Vision-Language-Action Policies

arXiv:2604.24622 · cs.CV, cs.AI · Submitted 2026-04-27 · Read on arXiv

cs.CV, cs.AI

Submitted: 2026-04-27

Updated: 2026-09-04

Comments: Accepted to ACM Multimedia (ACM MM) 2026 as an Oral Presentation

DOI: 10.1145/3767308.3836126

Code: https://github.com/EmbodiedAI-RoboTron/CF-VLA

License: http://creativecommons.org/licenses/by/4.0/

The gist: Flow-based vision-language-action (VLA) policies offer strong expressivity for action generation, but suffer from a fundamental inefficiency: multi-step inference is required to recover action

Terminology

Abstract

Flow-based vision-language-action (VLA) policies offer strong expressivity for action generation, but suffer from a fundamental inefficiency: multi-step inference is required to recover action structure from uninformative Gaussian noise, leading to a poor efficiency-quality trade-off under real-time constraints. We address this issue by rethinking the role of the starting point in generative action modeling. Instead of shortening the sampling trajectory, we propose CF-VLA, a coarse-to-fine two-stage formulation that restructures action generation into a coarse initialization step that constructs an action-aware starting point, followed by a single-step local refinement that corrects residual errors. Concretely, the coarse stage learns a conditional posterior over endpoint velocity to transform Gaussian noise into a structured initialization, while the fine stage performs a fixed-time refinement from this initialization. To stabilize training, we introduce a stepwise strategy that first learns a controlled coarse predictor and then performs joint optimization. Experiments on CALVIN and LIBERO show that our method establishes a strong efficiency-performance frontier under low-NFE (Number of Function Evaluations) regimes: it consistently outperforms existing NFE=2 methods, matches or surpasses the NFE=10 π 0.5 baseline on several metrics, reduces action sampling latency by 75.4%, and achieves the best average real-robot success rate of 83.0%, outperforming MIP by 19.5 points and π 0.5 by 4.0 points. These results suggest that structured, coarse-to-fine generation enables both strong performance and efficient inference. Our code is available at https://github.com/EmbodiedAI-RoboTron/CF-VLA.

Sources

Related papers