Reuse Before You Retrieve: Diagnosing Headroom and Complementarity for Test-Time Augmentation of Embodied Multimodal Policies

📄 arXiv: 2608.17484v1 📥 PDF

作者: Yuhwan Jeong, Kuk-Jin Yoon

分类: cs.RO

发布日期: 2026-08-18

备注: Accepted to ECCV 2026 workshop


💡 一句话要点

提出测试时增强策略以优化多模态政策的行为选择

🎯 匹配领域: 支柱九:具身大模型 (Embodied Foundation Models)

关键词: 多模态政策 测试时增强 行为选择 机器人控制 视觉-语言-动作

📋 核心要点

  1. 现有的冻结VLA政策在测试时缺乏有效的干预选择指导,导致潜在能力未被充分利用。
  2. 论文提出通过可恢复的潜力和检索互补性来评估政策的行为需求,从而优化测试时的增强策略。
  3. 实验表明,重试选择器在多个VLA政策上均能显著提升成功率,最高可达21.0个百分点,且在不同环境中保持有效。

📝 摘要(中文)

冻结的视觉-语言-动作(VLA)政策在测试时通过采样额外的政策行为或引入外部示范不断得到改善。然而,如何决定已部署政策实际需要何种干预的指导仍然不足。本文通过可恢复的潜力和检索互补性两个可测量因素,研究了这一决策过程。我们评估了在可重试或并行执行下的回合级重试选择器,并在多个冻结的VLA政策和环境中进行检索。实验结果显示,该选择器在LIBERO上显著恢复了潜在能力,成功率提升高达21.0个百分点,且在不同机器人和模拟器上均有效。检索在填补政策行为先验的差距时表现出不同的行为,结合选择时可获得进一步提升。这些结果为在测试时增强机会的表征提供了实证基础。

🔬 方法详解

问题定义:本文旨在解决冻结的VLA政策在测试时如何选择有效的干预措施的问题。现有方法未能充分利用政策的潜在能力,导致行为选择不够优化。

核心思路:论文提出通过分析可恢复的潜力和检索互补性来指导政策的增强策略。可恢复的潜力指的是政策内部可利用的行为,而检索互补性则评估外部行为先验的补充效果。

技术框架:整体架构包括回合级重试选择器和检索机制。重试选择器在可重试或并行执行的情况下,选择最优的行为进行执行,而检索机制则从多个冻结的VLA政策中获取外部示范。

关键创新:最重要的创新在于通过可测量的因素(可恢复的潜力和检索互补性)来系统性地评估和优化测试时的政策行为选择。这种方法与传统的单一干预策略有本质区别。

关键设计:在实验中,选择器的参数设置经过精细调整,以确保在不同环境下的有效性。同时,损失函数设计考虑了行为选择的多样性和成功率的提升,确保了政策的灵活性和适应性。

🖼️ 关键图片

img_0
img_1
img_2

📊 实验亮点

实验结果显示,回合级重试选择器在LIBERO测试中成功率提升高达21.0个百分点,且在不同机器人和模拟器上均表现出色。这表明该方法在恢复政策潜力和优化行为选择方面具有显著效果,尤其是在面对观察条件恶化时依然有效。

🎯 应用场景

该研究的潜在应用领域包括机器人控制、自动驾驶和智能助手等多模态交互系统。通过优化测试时的政策行为选择,可以显著提升系统的自主性和适应能力,进而推动智能系统在复杂环境中的应用。未来,该方法有望在更多实际场景中得到验证和推广。

📄 摘要(原文)

Frozen vision-language-action (VLA) policies are increasingly improved at test time by sampling additional policy behaviors or introducing external demonstrations. Yet there is little guidance for deciding which intervention a deployed policy actually needs. Additional sampling is useful only when better behavior already exists within the policy's stochastic rollouts and can be identified, whereas retrieval is most useful when the relevant action prior is not reliably represented by the policy. We study this decision through two measurable factors, recoverable headroom and retrieval complementarity, which characterize how much useful behavior is already available to recover and whether an external action prior fills a measurable gap. We evaluate an episode-level retry selector under retryable or parallel execution, together with retrieval across multiple frozen VLA policies and environments. The selector consistently recovers substantial latent capability across all tested VLA backbones on LIBERO, with gains of up to 21.0 success-rate points that closely track recoverable headroom. It also transfers to a different robot and simulator and remains effective under degraded observations, while experiments with autoregressive OpenVLA illustrate the distinction between available headroom and the ability to rank candidate rollouts. Retrieval behaves differently, improving the policy with the largest measured action-prior gap and providing further gains when combined with selection. Together, these results provide an empirical basis for characterizing test-time augmentation opportunities by separating capability that can be recovered from the frozen policy from behavioral priors that may need to be introduced externally.