用隐状态推理+提前退出，让视觉语言动作模型更快更稳。
Think Less, Act Early: Reinforced Latent Reasoning with Early Exit in Vision-Language-Action Models
- 将推理过程建模为不可见的隐变量序列，避免生成冗余文本。
- 在LIBERO上实现98.3%成功率，推理速度提升6倍。
- 适合需要高效长程决策的机器人任务场景。
现有视觉-语言-动作（VLA）模型多依赖显式的思维链（CoT）推理来连接感知与行动，虽有效但计算开销大且易在多步任务中传播错误。本文提出自适应可变对齐VLA（AVA-VLA），一种新型隐式推理框架，将推理视为一系列不可观测的隐变量序列，无需生成显式文本。然而，隐状态轨迹易受噪声干扰且与下游目标错位。为此，我们引入基于强化学习的去噪机制，将隐状态生成视为序列决策过程，通过任务级奖励优化推理路径。此外，集成提前退出策略，依据状态置信度自适应终止推理，实现深度与效率的动态平衡。在具身决策基准上的大量实验表明，AVA-VLA相比显式CoT方法实现6倍推理加速，同时在LIBERO上达到98.3%平均成功率，显著提升效率与长程稳定性。
原文摘要 · Abstract (English)
Existing Vision-Language-Action (VLA) models predominantly rely on explicit Chain-of-Thought (CoT) reasoning to bridge perception and action. While effective, this paradigm suffers from high computational costs and error propagation in multi-step tasks. In this paper, we propose Adaptive Variable Alignment VLA (AVA-VLA), a novel Latent Reasoning VLA framework that models reasoning as a sequence of unobservable latent variables, bypassing the need for explicit text generation. However, latent trajectories are inherently susceptible to noise interference and misalignment with downstream objectives. To address this, we introduce a Reinforcement Learning-based Denoising mechanism that treats latent state generation as a sequential decision process, optimizing reasoning trajectories via task-level rewards. Furthermore, we incorporate an Early-Exit Strategy that adaptively terminates reasoning based on state confidence, enabling a dynamic trade-off between depth and efficiency. Extensive experiments on embodied decision benchmarks demonstrate that AVA-VLA achieves a 6x inference speedup over explicit CoT methods while attaining a 98.3% average success rate on LIBERO, improving both efficiency and long-horizon stability over full-reasoning baselines.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。