SIR：面向可解释机器人学习的结构化图像表征
SIR: Structured Image Representations for Explainable Robot Learning
用场景图作为机器人策略的中间表征，把图像特征转成可稀疏化的图结构，既提升对干扰物的鲁棒性，又让决策过程可被逐节点审查；做视觉表征或 VLA 的研究者可借鉴这种「结构化中间表示 + 端到端稀疏化」的思路来替代黑箱图像嵌入。
原文依据 we introduce Structured Image Representation, a method that leverages Scene Graphs as an intermediate representation for robot policy learning
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论现有的基于学习视觉嵌入的机器人策略缺乏显式结构，且对视觉干扰敏感。因此，驱动其行为的表征往往不透明，使其决策过程难以解释。为解决这一问题，我们提出结构化图像表征（Structured Image Representation），一种利用场景图作为机器人策略学习中间表征的方法。我们的方法首先构建一个全连接图，以由 2D 或 3D 图像导出的特征作为初始节点表征。随后，一个模块以端到端方式学习对该图进行稀疏化，生成一个最小的、与任务相关的子图，并将其传递给动作生成模型。这一过程使我们的模型具有内在可解释性。在 RoboCasa 上的评估表明，我们的稀疏图策略平均成功率为 19.5%，优于基于图像的基线的 14.81%。我们还证明，基于图的表征对干扰物体显著更鲁棒，几乎不出现性能下降，而图像表征则不然。最重要的是，我们展示了学习到的稀疏图是进行内省的有力工具。通过分析模型的子图何时偏离人类预期，例如包含干扰节点或遗漏关键物体，我们成功揭示了数据集偏差，包括虚假相关和位置偏差。
英文原文摘要 1395 字符
Existing robot policies based on learned visual embeddings lack explicit structure and are sensitive to visual distractions.Thus, the representations that drive their behaviour are often opaque, making their decision-making process difficult to interpret.To address this, we introduce Structured Image Representation, a method that leverages Scene Graphs as an intermediate representation for robot policy learning.Our approach first constructs a fully connected graph, using 2D or 3D image-derived features as initial node representations. Then, a module learns to sparsify this graph end-to-end, creating a minimal, task-relevant sub-graph that is passed to the action generation model.This process makes our model intrinsically explainable.Evaluations on RoboCasa show that our sparse graph policies outperform image-based baselines on average with 19.5% vs 14.81% success rate.We also demonstrate that our graph-based representations are significantly more robust to distractor objects, showing almost no performance degradation, as opposed to image representations.Most importantly, we show that the learned sparse graphs are a powerful tool for introspection.By analysing when the model's sub-graph deviates from human expectation, such as by including distractor nodes or omitting key objects, we successfully uncover dataset biases, including spurious correlations and positional biases.