[论文] What Matters in Quality-Aware VLA Pretraining: A Benchmark and Empiric...

论文概要

研究领域: ML 作者: Mingxuan Yan, Litian Gong, Dihong Huang, Zhixu Li, Tianyu Zhang, Wenqian Zhang, Hefeifei Jiang, Peihao Li, Jian Zhang, Mengfei Zhao, Yikai Tang, Hai Zhai, Zehao Wang, Ruijian Liang, Yanjia Huang, Lin Shao, Changliu Liu, Kaiyu Hang, Zhiwen Fan, Masayoshi Tomizuka, Jianfei Yang, Jiachen Li 发布时间: 2026-10-13 arXiv: 2610.18869

中文摘要

在视觉-语言-动作（VLA）预训练中，应如何利用演示质量来提升目标任务微调后的性能？我们使用 5,000 条自然次优的人类演示和 24 个目标任务，在奖励定义和标签粒度维度上比较了加权、过滤和条件化三种策略。在每个 VLA 内，我们匹配了演示数据和训练预算。额外比较了预训练数据规模和品质标签在微调及推理阶段的作用。横跨三个 VLA，我们发现：感知感知的回合条件化优于普通行为克隆；条件化和过滤分别更偏好回合级和块级标签（存在模型依赖的反转）；条件化预训练后使用恒定标签微调可提升平均成功率（无论推理时是否带标签）。我们推出 AXIS-Bench 支持这些受控比较，基于众包仿真演示构建，提供记录的物体和接触状态用于在共享轨迹上重算质量标签。可重放轨迹支持工作空间对齐渲染和用户定义的跨域目标任务。真实世界实验中，仿真预训练将成功率从 9.6% 提升至 31.7%。

原文摘要

How should demonstration quality be used during vision-language-action (VLA) pretraining to improve performance after target-task fine-tuning? We compare weighting, filtering, and conditioning across reward definitions and label granularities using 5,000 naturally suboptimal human demonstrations and 24 target tasks. Within each VLA, we match demonstrations and training budgets. Additional comparisons examine pretraining data size and the roles of quality tags during fine-tuning and inference. Across three VLAs, we find that approach-aware episode conditioning improves mean success over plain behavioral cloning; conditioning and filtering favor episode and chunk labels, respectively, on average, with model-dependent reversals; and constant-tag fine-tuning improves mean success after conditi...

#论文 #arXiv #ML #小凯