论文/arXiv 2607.24538·3 个月前

NEO：一次 NeRF，多次编辑，用于连续物体操作

NEO面向连续物体操作，用语言引导NeRF编辑，并在场景编辑、物体移除和pick-and-place实验中验证。

AI 整理

• 数据与采集
• 端侧部署
• 评测与安全

原文链接：NEO: NeRF It Once, Edit It Many Times for Continuous Object Manipulation（arXiv 2607.24538 · 发布日期：2026-07-27）

NEO: NeRF It Once, Edit It Many Times for Continuous Object Manipulation

NEO面向连续物体操作，用语言引导NeRF编辑，并在场景编辑、物体移除和pick-and-place实验中验证。

核心要点

• 针对机器人操作中的NeRF场景编辑，目标是在动作执行前连贯建模未来场景状态。
• 提出语言引导物体移除，结合神经场重采样与多视图一致的渐进式修复。
• 通过知识蒸馏直接编辑NeRF权重，以教师-学生模型组合原始与编辑NeRF。
• 构建NEO-Dataset用于定量评估；方法优于现有基线，编辑更视觉连贯、几何一致。

验证范围与局限：摘要仅说明在场景编辑、物体移除和pick-and-place机器人实验中验证。

英文摘要 (Abstract)

In this paper, we present NEO, a unified framework providing language-guided NeRF editing for robotic manipulation. Our paper introduces (i) a language-guided object removal that combines neural field resampling with multiview-consistent progressive inpainting, (ii) a direct NeRF weight editing method utilizing knowledge distillation, composing original and edited NeRFs via a teacher-student model, enabling coherent modeling of future scene states before a robot executes an action, and (iii) the first benchmark (NEO-Dataset) for quantitatively evaluating NeRF scene editing methods suitable for robot manipulation. We show that our approach outperforms state-of-the-art baselines in scene editing tasks, including object removal and pick-and-place robotic experiments, yielding visually coherent and geometrically consistent edits that reduce artifacts commonly introduced by prior methods.

作者：Mikołaj Zieliński, David Hall, Dominik Belter, Peyman Moghadam

来源

本页事实对应的公开页面，以原文为准。

1. arXiv 2607.24538：NEO: NeRF It Once, Edit It Many Times for Continuous Object Manipulation访问 2026-10-21