StructRL: Structured Action-Space Exploration for Flow-Based VLAs

2026-08-15

AI Summary

This study addresses the inefficient exploration caused by the dilution of structured noise in online reinforcement learning for streaming Vision-Language-Action (VLA) models. We propose StructRL, a novel method that pioneers a structured exploration paradigm within the action space. By integrating deterministic ODE decoding, action-space noise injection, and a terminal-step replay mechanism, StructRL effectively decouples exploration from denoising, thereby preventing intermediate steps from attenuating exploration signals. Experimental results demonstrate that StructRL significantly enhances both exploration efficiency and out-of-distribution generalization across simulated and real-world robotic tasks, outperforming existing baselines. Consequently, this work establishes a new pathway for the efficient and stable training of streaming VLA systems.

Abstract

Flow-based Vision-Language-Action (VLA) models are now widely used for continuous robotic manipulation, and online reinforcement learning (RL) is emerging as a key technique for adapting them to new tasks. Existing RL methods typically inject stochasticity inside the denoising chain, often through isotropic or temporally independent noise. However, effective robot exploration calls for structured noise: temporally smooth and scaled differently across action groups. We show that simply switching the in-chain noise to a structured form does not suffice: noise added at an intermediate flow time can be weakened by the remaining denoising steps before execution, a phenomenon we call \emph{Structured Noise Dilution}. We propose \textbf{StructRL}, which avoids dilution by relocating policy stochasticity to the action space via three coupled choices: (i) a deterministic ODE decoder, (ii) structured noise injected directly in the action space, and (iii) last-step replay, where policy-gradient updates avoid assigning likelihoods to intermediate denoising states. This keeps structured exploration tied to the executed action while providing a tractable training signal for the flow decoder. Across three flow-based VLA models on multiple simulated manipulation benchmarks and two real-world tasks, StructRL improves exploration efficiency and OOD performance over prior in-chain baselines, demonstrating the effectiveness of structured action-space exploration for adapting flow-based VLA with RL. \textbf{Project page:} https://flyfaerss.github.io/structrl/

Problem

• Flow-based VLA
• Online Reinforcement Learning
• Structured Noise Dilution
• Action-Space Exploration
• Robotic Manipulation

Innovation

• Structured Action-Space Exploration
• Flow-Based VLA
• Structured Noise Dilution
• Deterministic ODE Decoder
• Last-Step Replay