面向视觉-语言-动作模型的延迟反馈测试时扰动调优
Test-Time Perturbation Tuning with Delayed Feedback for Vision-Language-Action Models
对做视觉-语言-动作模型的研究者，这篇给出了一个无需微调基座模型、无需额外验证器的测试时自适应流程：用基于不确定性的数据增强加动作投票抑制动作与实体间的伪相关，再用延迟反馈驱动的轻量扰动模块校正动作 logits 的过度自信，可直接迁移到机器人操作策略的部署阶段。
原文依据 a verifier-free test-time adaptation framework that improves decision performance without fine-tuning the base model
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论视觉-语言-动作模型（VLA）在序列决策中取得了显著性能，但仍对细微的环境变化（例如物体位姿的小幅改变）十分脆弱。我们将这种脆弱性归因于轨迹过拟合，即 VLA 过度关注动作与实体之间的伪相关，进而复现记忆中的动作模式。为此，我们提出延迟反馈扰动学习（PDF），这是一个无需验证器的测试时自适应框架，可在不微调基座模型的前提下提升决策性能。PDF 通过基于不确定性的数据增强与动作投票来缓解伪相关，并由一个自适应调度器分配增强预算以平衡性能与效率。为进一步提升稳定性，PDF 学习一个轻量扰动模块，在延迟反馈的引导下回溯性地调整动作 logits，从而纠正过度自信问题。在 LIBERO（成功率 +7.4%）与 Atari（人类归一化得分 +10.3）上的实验表明，PDF 在任务成功率上相较原始 VLA 以及采用测试时自适应的 VLA 均取得稳定提升，为多模态决策智能体中可靠的测试时自适应建立了一条切实可行的路径。
英文原文摘要 1218 字符
Vision-Language-Action models (VLAs) achieve remarkable performance in sequential decision-making but remain fragile to subtle environmental shifts, such as small changes in object pose. We attribute this brittleness to trajectory overfitting, where VLAs over-attend to the spurious correlation between actions and entities, then reproduce memorized action patterns. We propose Perturbation learning with Delayed Feedback (PDF), a verifier-free test-time adaptation framework that improves decision performance without fine-tuning the base model. PDF mitigates the spurious correlation through uncertainty-based data augmentation and action voting, while an adaptive scheduler allocates augmentation budgets to balance performance and efficiency. To further improve stability, PDF learns a lightweight perturbation module that retrospectively adjusts action logits guided by delayed feedback, correcting overconfidence issue. Experiments on LIBERO (+7.4% success rate) and Atari (+10.3 human normalized score) demonstrate consistent gains of PDF in task success over vanilla VLA and VLA with test-time adaptation, establishing a practical path toward reliable test-time adaptation in multimodal decision-making agents.