ProbeFlow achieves 14.8x faster action decoding in Vision-Language-Action (VLA) models without any retraining.
March 19, 2026
Original Paper
ProbeFlow: Training-Free Adaptive Flow Matching for Vision-Language-Action Models
arXiv · 2603.17850
The Takeaway
By using a simple geometric probe to determine trajectory complexity, it allows robotic controllers to skip redundant network evaluations on simple tasks. This effectively removes the inference latency bottleneck that often prevents diffusion-based models from achieving real-time physical control.
From the abstract
Recent Vision-Language-Action (VLA) models equipped with Flow Matching (FM) action heads achieve state-of-the-art performance in complex robot manipulation. However, the multi-step iterative ODE solving required by FM introduces inference latency that precludes responsive physical control. While current acceleration efforts optimize the Vision-Language Model (VLM) backbone, the action head bottleneck remains overlooked. To address this, we propose ProbeFlow, a training-free adaptive inference fr