RoboDrop: Curating VLA Post-Training Data via Local Gradient Compatibility

📄 arXiv: 2609.10021v1 📥 PDF

作者: Runze Xu, Yuanfan Xu, Cuijie Xu, Shuang Dai, Yining Li, Yu Wang, Jincheng Yu

分类: cs.RO

发布日期: 2026-09-09


💡 一句话要点

提出RoboDrop以解决VLA模型后训练数据质量问题

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

关键词: 视觉-语言-动作 后训练 数据策划 局部梯度兼容性 机器人学习 数据清理 策略优化

📋 核心要点

  1. 现有的后训练方法对数据质量敏感,手动检查成本高且效率低,难以处理异构错误。
  2. RoboDrop通过局部梯度兼容性审计监督,在线评分候选样本,自动化数据清理过程。
  3. 在多种实验设置下,RoboDrop能够更准确地区分不可靠的演示,显著提升后训练策略的成功率。

📝 摘要(中文)

视觉-语言-动作(VLA)模型通过大规模预训练获得广泛的泛化能力,但在适应新任务和机器人形态时仍需对新收集的数据进行后训练。现有方法在数据质量上存在不足,手动检查成本高,且现有数据清理方法通常针对特定的错误类型。为了解决这些挑战,本文提出了RoboDrop,一个通过局部梯度兼容性审计监督的数据策划框架。RoboDrop在一个热身运行中在线评分候选样本,并通过与任务语义和视觉匹配的验证样本的梯度比较来评估其对后训练性能的影响。实验结果表明,RoboDrop在区分不可靠演示方面优于现有方法,后训练的策略性能显著提升,真实机器人实验成功率从35.0%提高到67.5%。

🔬 方法详解

问题定义:本文旨在解决在机器人后训练过程中,由于数据质量不高(如执行错误、传感器漂移等)导致的性能下降问题。现有方法往往无法有效处理这些异构错误,手动检查的成本也非常高。

核心思路:RoboDrop的核心思路是利用局部梯度兼容性作为监督审计的代理,通过比较候选样本的梯度与任务语义和视觉匹配样本的梯度,来评估其对后训练性能的影响。

技术框架:RoboDrop的整体流程包括一个热身运行阶段,在此阶段对每个候选样本进行在线评分,随后将样本分数在回合级别进行聚合,最终通过简单的自动后处理规则做出过滤决策。

关键创新:RoboDrop的主要创新在于引入了训练轨迹感知的上下文条件监督审计方法,这种方法能够更有效地识别不可靠的演示,与传统的针对特定错误类型的数据清理方法本质上不同。

关键设计:在实现上,RoboDrop设计了一个评分机制,通过比较梯度来评估样本的可靠性,且在后处理阶段采用了简单的规则来决定样本的保留与否。

🖼️ 关键图片

fig_0
fig_1
fig_2

📊 实验亮点

实验结果显示,RoboDrop在控制观察-动作的错误、自然次优演示以及真实机器人数据集上表现优异,能够更准确地区分不可靠的演示。后训练后,策略的成功率从35.0%显著提升至67.5%,展示了其在提升后训练效果方面的强大能力。

🎯 应用场景

RoboDrop的研究成果在机器人学习和自动化领域具有广泛的应用潜力,尤其是在需要高质量数据进行后训练的任务中。通过提高数据质量,RoboDrop能够帮助机器人更好地适应新环境和任务,提升其自主决策能力,未来可能在智能制造、服务机器人等领域产生深远影响。

📄 摘要(原文)

Vision--language--action (VLA) models acquire broad generalization through large-scale pretraining, yet adapting them to a new task and robot embodiment still requires post-training on newly collected data. Unlike pretraining, post-training targets task- and embodiment-specific adaptation, making it particularly sensitive to data quality. In practice, collected robot datasets often contain heterogeneous errors, including execution mistakes, sensor drift, and timestamp misalignment, which can impair post-training and policy performance. Manual inspection is costly, while existing data-cleaning methods are typically tailored to particular corruption types. To address these challenges, we introduce \textsc{RoboDrop}, a data-curation framework that audits supervision using local gradient compatibility measured along the training trajectory as a proxy for its effect on post-training performance. During a one-epoch warm-up run, RoboDrop scores each candidate sample online by comparing its gradient with those of task-semantic and visually matched validation samples. The resulting sample scores are aggregated at the episode level, and a simple automatic post-processing rule converts them into filtering decisions. We evaluate RoboDrop on controlled observation--action corruptions, naturally suboptimal demonstrations in simulation, and real-robot datasets containing non-expert collection errors. Across these settings, RoboDrop more accurately distinguishes unreliable demonstrations than prior methods, while post-training on the curated data consistently yields stronger downstream policies, with average real-robot rollout success rising from $35.0\%$ to $67.5\%$. These results establish training-trajectory-aware, context-conditioned supervision auditing as an effective approach to robust VLA post-training.