Kintsugi-VLA: Turning Failed Robot Rollouts into Recovery Data through Interventional Recoverability
作者: Ivan Snegirev, Elizaveta Semenyakina, Dmitrii Maliukov, Miguel Altamirano Cabrera, Dzmitry Tsetserukou
分类: cs.RO
发布日期: 2026-09-25
备注: 8 pages, 7 figures, 5 tables (applied on ICRA2027)
💡 一句话要点
提出Kintsugi-VLA框架以将失败机器人轨迹转化为恢复数据
🎯 匹配领域: 支柱一：机器人控制 (Robot Control) 支柱九：具身大模型 (Embodied Foundation Models)
关键词: 机器人恢复 视觉-语言-动作 干预可恢复性 模拟学习 数据增强
📋 核心要点
- 现有的视觉-语言-动作策略训练方法通常忽视失败轨迹，导致无法有效学习恢复策略。
- Kintsugi-VLA框架通过将失败轨迹转化为合成恢复数据，利用状态恢复和干预可恢复性来优化恢复学习。
- 在Franka操作任务中，针对性恢复数据的成功率分别提高了5.8和6.7个百分点，展示了该方法的有效性。
📝 摘要（中文）
模拟环境使得通过特权专家生成视觉演示来训练视觉-语言-动作策略成为可能。然而，现有方法通常只保留成功的演示，忽略了失败轨迹中蕴含的恢复学习信息。本文提出Kintsugi-VLA框架，通过精确状态恢复和分支模拟，将失败轨迹转化为有针对性的合成恢复数据。我们定义了干预可恢复性，并通过自适应蒙特卡洛方法进行估计，识别出低恢复性边界，从而选择信息丰富的恢复起始状态。在模拟的Franka操作任务中，针对性恢复数据的成功率显著高于均匀采样。
🔬 方法详解
问题定义：现有方法在训练视觉-语言-动作策略时，通常只保留成功的演示，导致失败轨迹中的重要恢复信息被丢弃，无法有效学习如何应对非正常状态。
核心思路：Kintsugi-VLA框架通过将失败轨迹转化为合成恢复数据，利用干预可恢复性来评估在特定状态下完成任务的概率，从而优化恢复策略的学习。
技术框架：该框架主要包括状态恢复模块、干预可恢复性估计模块和数据选择模块。首先，通过状态恢复将模拟器恢复到失败状态；然后，计算干预可恢复性以评估恢复成功的可能性；最后，选择信息丰富的起始状态进行恢复训练。
关键创新：该研究的创新点在于引入了干预可恢复性这一概念，并通过自适应蒙特卡洛方法进行估计，能够有效识别低恢复性边界，从而优化恢复数据的选择。
关键设计：在技术细节上，采用了点估计的威尔逊区间来评估恢复性，并通过动态调整恢复起始状态来提高训练数据的有效性。
🖼️ 关键图片
📊 实验亮点
在模拟Franka操作任务中，Kintsugi-VLA框架的针对性恢复数据成功率达34.6%和38.4%，分别比均匀采样提高了5.8和6.7个百分点。这一结果表明，该方法在不同条件下均能有效提升恢复成功率，展示了其强大的实用性。
🎯 应用场景
Kintsugi-VLA框架在机器人操作和自动化领域具有广泛的应用潜力，特别是在需要高可靠性的任务中。通过有效利用失败轨迹，该方法可以提升机器人在复杂环境中的适应能力和恢复能力，未来可能在智能制造、服务机器人等领域发挥重要作用。
📄 摘要（原文）
Simulation enables scalable training of Vision-Language-Action policies by using privileged experts to generate visual demonstrations without requiring every trajectory to be collected through manual teleoperation. However, such pipelines typically retain successful demonstrations while failed rollouts are discarded, even though they expose precisely the off-nominal states from which recovery must be learned. We introduce Kintsugi-VLA, a framework for converting failed rollouts into targeted synthetic recovery data by exploiting exact state restoration and branching in simulation. For a fixed privileged expert, we define interventional recoverability as the probability of completing the original task after the simulator is restored to a given state, estimate it using adaptive Monte Carlo continuations with pointwise Wilson intervals, and characterize its non-monotonic evolution along failed trajectories. These estimates identify an observed terminal low-recoverability frontier-the point after which measured recoverability remains below a threshold-which is then used to select informative recovery starting states. In a simulated Franka manipulation task, targeted recovery data yield aggregate SmolVLA recovery success of 34.6\% and 38.4\% under difficulty- and frame-budget matching, respectively, 5.8 and 6.7 percentage points above uniform sampling within the same recovery window. The same ordering is observed under disturbed end-to-end execution and shifted clutter and physics conditions, while clean-task success decreases from 76.8\% to 74.7\%. Kintsugi-VLA demonstrates how failed simulator rollouts can be transformed from discarded experience into structured recovery-training data through direct interventional measurement.