arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2503.07557cs.RO
收藏

AutoSpatial: 通过高效空间推理学习实现社交机器人导航的视觉-语言推理

AutoSpatial: Visual-Language Reasoning for Social Robot Navigation through Efficient Spatial Reasoning Learning

George Mason University(乔治·马歇尔大学);University of Maryland, College Park(马里兰大学学院公园分校)

Yangzhe Kong, Daeun Song, Jing Liang, Dinesh Manocha, Ziyu Yao, Xuesu Xiao

AI总结:

本文提出AutoSpatial方法,通过结构化空间接地和大规模VQA对自动生成标签,提升VLMs在社交导航中的空间推理能力,实现更准确的空间感知、运动预测、推理、行动和解释。

摘要:

我们提出了一种新颖的方法AutoSpatial,一种具有结构化空间接地的高效方法,以增强VLMs的空间推理能力。通过结合最小的手动监督和大规模视觉问答(VQA)对自动生成标签,我们的方法解决了VLMs在社交导航任务中空间理解有限的挑战。在训练过程中应用分层两轮VQA策略,AutoSpatial实现了对场景的全局和详细理解,显示出比其他最先进方法更准确的空间感知、运动预测、链式思维(CoT)推理、最终行动和解释。这五个组成部分对于全面的社会导航推理至关重要。我们的方法通过专家系统(GPT-4o、Gemini 2.0 Flash和Claude 3.5 Sonnet)提供的交叉验证分数和人类评估者分配的相对排名,进行了评估,以比较模型在四个关键方面的表现。增强的空间推理能力使AutoSpatial在感知与预测(最高提升10.71%)、推理(最高提升16.26%)、行动(最高提升20.50%)和解释(最高提升18.73%)方面相比仅使用手动标注数据训练的基线模型显示出显著改进。

英文摘要:

We present a novel method, AutoSpatial, an efficient approach with structured spatial grounding to enhance VLMs' spatial reasoning. By combining minimal manual supervision with large-scale Visual Question-Answering (VQA) pairs auto-labeling, our approach tackles the challenge of VLMs' limited spatial understanding in social navigation tasks. By applying a hierarchical two-round VQA strategy during training, AutoSpatial achieves both global and detailed understanding of scenarios, demonstrating more accurate spatial perception, movement prediction, Chain of Thought (CoT) reasoning, final action, and explanation compared to other SOTA approaches. These five components are essential for comprehensive social navigation reasoning. Our approach was evaluated using both expert systems (GPT-4o, Gemini 2.0 Flash, and Claude 3.5 Sonnet) that provided cross-validation scores and human evaluators who assigned relative rankings to compare model performances across four key aspects. Augmented by the enhanced spatial reasoning capabilities, AutoSpatial demonstrates substantial improvements by averaged cross-validation score from expert systems in: perception & prediction (up to 10.71%), reasoning (up to 16.26%), action (up to 20.50%), and explanation (up to 18.73%) compared to baseline models trained only on manually annotated data.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑