用注意力机制从卷积特征中提取关键视觉局部信息，提升机器人控制效率。
Localized Visual Feature Aggregation via Focus Pooling for Visuomotor Policies
- 设计注意力池化模块FocusPool，动态聚合与机器人状态相关的局部视觉特征。
- 在仿真和真实场景中，成功率比传统方法高36.2%~41.2%，仅训练5.8%参数。
- 适合需要高效视觉感知的机器人控制任务，尤其对计算资源有限场景友好。
聚焦空间上局部、与控制相关的视觉线索已被证明可通过减少对任务无关视觉变化的建模，提升视觉-运动策略的数据效率。现有方法通常通过输入预处理实现聚焦，例如裁剪图像或点云中的控制中心或物体中心区域。然而，尚未深入探索这些局部特征是否可直接从常用的卷积神经网络（CNN）编码特征中提取。本文表明，中间层的CNN特征保留了与控制相关的局部视觉上下文，但现有池化方法无法有效聚合。为此，我们提出FocusPool——一种注意力池化模块，根据特征与当前机器人本体感受状态的相关性，选择性地聚合中间视觉特征。生成的池化表示捕捉了任务进展中的控制相关局部信息，并直接用于策略学习。在仿真和真实世界实验中，FocusPool相较于普通池化和显式局部聚焦方法，成功率分别提升了36.2%和41.2%，且仅需训练5.8%的编码器参数。
原文摘要 · Abstract (English)
Focusing on spatially localized, control-relevant visual cues has been shown to improve data efficiency in visuomotor policies by reducing the need to model task-irrelevant visual variation. Existing methods often impose this focus through input preprocessing, such as cropping control- or object-centric regions in RGB images or point-clouds. However, it remains underexplored whether such localized features can be exposed directly from commonly used convolutional neural network (CNN) encoded features. In this paper, we show that intermediate CNN features preserve localized visual context for control, but existing pooling methods fail to aggregate it effectively. We introduce FocusPool, an attention pooling module that selectively aggregates intermediate visual features according to their relevance to the robot's current proprioceptive context. The resulting pooled representation captures task-progressive, control-relevant local information and is used directly for policy learning. Across simulation and real-world experiments, FocusPool improves policy success rates over pooling and explicit local focus methods by 36.2% and 41.2%, with training only 5.8% of encoder parameters.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。