[论文] ProtoVis-Nav: Prototypical Visual Imagination for Visual-Spatial Align...

论文概要

研究领域: CV 作者: Guangyuan Sui, Yunzhong Hou, Chi Harold Liu 发布时间: 2026-11-12 arXiv: 2611.17330

中文摘要

视觉语言导航（VLN）对无人机而言极具挑战性——3D 运动预测中的微小误差就可能导致导航失败甚至碰撞。现有方法无法在 3D 环境中预判执行结果，往往难以从语言指令推断出准确的导航路径点。为此我们提出原型视觉想象框架 ProtoVis-Nav：强制模型为下一个预期 UAV 动作想象一个与结果关联的局部视觉表示。具体而言，模型预测特征原型，用其线性组合估计目标区域的视觉特征。这样，VLN 模型先"想象"下一个目标位置的局部视觉外观，再以预测原型为上下文生成最终动作，促进视觉上下文与路径点输出的对齐。大量实验验证了 ProtoVis-Nav 的有效性，相比现有基线有显著提升。

原文摘要

Vision language navigation (VLN) is very challenging for Unmanned Aerial Vehicles (UAVs) since small errors in 3D motion prediction can lead to navigation errors and collisions. Unable to fathom execution outcomes in 3D environments, existing methods often struggle to predict accurate navigation waypoints from language instructions. To solve this, we introduce the Prototypical Visual Imagination framework, shortened as ProtoVis-Nav, which forces the model to imagine an outcome-associated local visual representation for the intended next UAV movement. Specifically, we predict feature prototypes and use their linear combination to estimate the visual feature of the target region. In this manner, the VLN model first imagines the local visual representation of the next intended location, and t...

#论文 #arXiv #CV #小凯