Does Imitation Learning Preserve Temporal Robustness in Dexterous Manipulation? An Expert-Learner Comparison Across Task Execution Speeds

📄 arXiv: 2609.01453v1 📥 PDF

作者: Clinton Enwerem, John S. Baras, Calin Belta

分类: cs.RO, cs.LG

发布日期: 2026-09-01

备注: 19 pages, 10 figures, and 9 tables. Code, data, and evaluation scripts are available at https://github.com/coenwerem/parcelstow

🔗 代码/项目: GITHUB

💡 一句话要点

比较模仿学习与专家在灵巧操作中的时间鲁棒性

🎯 匹配领域: 支柱一：机器人控制 (Robot Control) 支柱二：RL算法与架构 (RL & Architecture)

关键词: 模仿学习 灵巧操作 时间鲁棒性 机器人技术 任务执行速度 专家学习 性能评估

📋 核心要点

  1. 现有的模仿学习方法通常忽视了在不同任务执行速度下的鲁棒性评估，导致对学习者性能的理解不全面。
  2. 论文通过在ParcelStow任务中比较专家与学习者的表现，探讨了模仿学习在时间鲁棒性方面的保留程度。
  3. 实验结果表明，尽管名义速度下成功率相同，但在加速条件下，学习者的表现显著下降，专家的鲁棒性更强。

📝 摘要（中文）

模仿学习的灵巧操作策略通常评估其对场景、物体或指令变化的鲁棒性，但对任务执行速度的表现较少研究。本文比较了专家与学习者在相同任务条件下的表现，使用ParcelStow任务进行评估。结果显示，尽管在名义速度下两者均实现100%的任务成功率，但在加速执行时，专家的成功率显著高于学习者。具体而言，专家在最大速度下的成功率为84%，而学习者为53%。这表明，名义任务成功并不意味着在执行速度上的专家表现得以保留。

🔬 方法详解

问题定义：本文旨在解决模仿学习在灵巧操作中对时间鲁棒性评估不足的问题。现有方法通常只关注场景和对象变化，而忽略了任务执行速度对性能的影响。

核心思路：通过在ParcelStow任务中比较专家与学习者的表现，评估模仿学习策略在不同执行速度下的鲁棒性，揭示其潜在的性能差距。

技术框架：研究采用了ParcelStow任务，机器人需获取、重新定向并插入包裹。实验中设置了不同的速度加速因子，并在相同的初始条件下进行比较。

关键创新：论文的创新在于系统性地评估了模仿学习策略在时间维度上的鲁棒性，揭示了名义成功率与实际执行能力之间的差距。

关键设计：实验中使用了脚本专家和基于专家演示训练的Action Chunking with Transformers (ACT)策略，分析了不同参数初始化对学习者性能的影响。

🖼️ 关键图片

img_0
img_1
img_2

📊 实验亮点

实验结果显示，尽管专家和学习者在名义速度下均实现100%的任务成功率，但在最大速度下，专家成功率为84%，而学习者仅为53%。这表明模仿学习策略在速度变化下的鲁棒性显著低于专家。

🎯 应用场景

该研究的潜在应用领域包括机器人抓取与操作、自动化仓储系统以及人机协作等。通过提高模仿学习策略在不同执行速度下的鲁棒性，可以增强机器人在复杂环境中的适应能力，提升其实际应用价值。

📄 摘要（原文）

Dexterous manipulation policies learned by imitation are typically evaluated for robustness to variation in scenes, objects, or instructions, but their performance across task execution speeds is less often examined. This leaves open how much temporal robustness a learner retains relative to the expert it imitates. We compare an expert and learner under the same task conditions, initial-condition draws, and speedup factors. We instantiate the evaluation in ParcelStow, a contact-rich task in which the robot acquires, reorients, and inserts a parcel. The demonstrations span the speedup range for the manipulation phases after parcel acquisition. A scripted expert and an Action Chunking with Transformers (ACT) policy trained from the expert's demonstrations both achieve 100 percent task success at nominal speed. Their success rates diverge within the demonstrated range: at its maximum, expert success is 84 percent and ACT success is 53 percent. Two ACT policies with different parameter initializations show similar degradation, decreasing by 34 and 48 percentage points from nominal speed to the maximum demonstrated speed, compared with 16 points for the expert. Stage-level analysis shows that 35 of ACT's 47 failures at the maximum demonstrated speed are insertion misalignments. Under the relative-motion handoff, every ACT acquisition retains the parcel through reorientation and transfer in free space, but only 64 percent complete the overall task, compared with 95 percent after expert acquisition. Across all evaluated policies and speeds, none of the 414 acquisitions without force closure completes the task. Equal nominal task success therefore does not imply preservation of expert performance across execution speeds. Code, data, and evaluation scripts are available at https://github.com/coenwerem/parcelstow.