超越成功：通过即时干预从混合质量数据中精炼优雅的机器人操作
Beyond Success: Refining Elegant Robot Manipulation from Mixed-Quality Data via Just-in-Time Intervention
提出一种与基础 VLA 策略解耦的「优雅性评论家」加推理时按需干预机制：用离线校准 Q 学习估计候选动作的执行质量，仅在评论家置信度低的决策关键点介入。做 VLA 策略后处理、动作质量重排序或推理时对齐的视觉研究者可直接复用该「评论家+选择性干预」范式，无需重训大模型。
原文依据 a Just-in-Time Intervention (JITI) mechanism monitors critic confidence and intervenes only at decision-critical moments
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（VLA）模型已在通用机器人操作上取得显著进展，但其学到的策略往往表现出参差不齐的执行质量。我们将这种波动归因于人类演示数据的混合质量特性——支配动作应如何执行的隐含原则只被部分满足。为应对这一挑战，我们提出了 LIBERO-Elegant 基准，并给出评估执行质量的明确标准。基于这些标准，我们设计了一个解耦的精炼框架，在不修改、不重训基础 VLA 策略的前提下提升执行质量。我们将「优雅执行」形式化为对隐式任务约束（ITC）的满足，并通过离线校准 Q 学习训练一个优雅性评论家，用以估计候选动作的期望质量。在推理时，即时干预（JITI）机制监测评论家的置信度，仅在决策关键时刻介入，从而提供有选择、按需的精炼。在 LIBERO-Elegant 与真实世界操作任务上的实验表明，所学的优雅性评论家能显著提升执行质量，即使在未见过的任务上亦然。所提模型使机器人控制不仅关注任务是否成功，也关注任务是如何被执行的。
英文原文摘要 1316 字符
Vision-Language-Action (VLA) models have enabled notable progress in general-purpose robotic manipulation, yet their learned policies often exhibit variable execution quality. We attribute this variability to the mixed-quality nature of human demonstrations, where the implicit principles that govern how actions should be carried out are only partially satisfied. To address this challenge, we introduce the LIBERO-Elegant benchmark with explicit criteria for evaluating execution quality. Using these criteria, we develop a decoupled refinement framework that improves execution quality without modifying or retraining the base VLA policy. We formalize Elegant Execution as the satisfaction of Implicit Task Constraints (ITCs) and train an Elegance Critic via offline Calibrated Q-Learning to estimate the expected quality of candidate actions. At inference time, a Just-in-Time Intervention (JITI) mechanism monitors critic confidence and intervenes only at decision-critical moments, providing selective, on-demand refinement. Experiments on LIBERO-Elegant and real-world manipulation tasks show that the learned Elegance Critic substantially improves execution quality, even on unseen tasks. The proposed model enables robotic control that values not only whether tasks succeed, but also how they are performed.