[论文] Scale Up Strategically: Learning Compositional Generalization via Bias...

论文概要

研究领域: CV 作者: Yu Qi, Zhang Ye, Xinyi Xu 发布时间: 2026-07-25 arXiv: 2507.20476

中文摘要

组合泛化对于机器人遵循多样化指令至关重要。然而，已知预训练策略会走捷径，依赖显著线索而非将语言落地。我们引入了一个诊断框架，将这一失败定位到单个指令因素，例如颜色、动词、物体、大小和空间属性等可重用的语义组件。我们的框架将指令因素偏置形式化，即微调策略过度依赖主导因素作为捷径的倾向，并通过两个指标量化：因素主导率（FDR），捕获因素之间的成对偏置；因素主导层级（FDH），将这些聚合成全局排序。在六个基础策略上的评估揭示了广泛一致的排序，即颜色 ≥ 物体 ≥ 空间 ≥ 动词 ≥ 大小，其中颜色占主导，动词和大小最缺乏grounding。我们进一步表明该诊断是可操作的：一种偏置感知的数据收集策略，将固定预算重新分配给 grounding 不足的因素，在模拟和真实机器人上均优于基线，使用一半的演示次数，从而实现更高效采样和更可泛化的策略学习。

原文摘要

Compositional generalization is essential for robot to follow diverse instructions. However, pretrained policies are known to take shortcuts, deferring to salient cues rather than grounding language. We introduce a diagnostic framework that localizes this failure to individual instruction factors, e.g., reusable semantic components such as color, verb, object, size, and spatial attribute. Our framework formalizes instruction factor bias, the tendency of fine-tuned policies to over-rely on dominant factors as shortcuts, and quantifies it through two metrics: Factor Dominance Rate (FDR), capturing pairwise bias between factors, and Factor Dominance Hierarchy (FDH), aggregating these into a global ranking. Evaluation on six foundation policies reveals broadly consistent ordering, i.e., color ...

--- *自动采集于 2026-07-26*

#论文 #arXiv #CV #小凯