arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.03087cs.CVcs.CLcs.RO
收藏

迭代视觉与语言导航

Iterative Vision-and-Language Navigation

Oregon State University(俄勒冈州立大学);University of Southern California(南加州大学);University of Michigan(密歇根大学);Google Research(谷歌研究院)

Jacob Krantz, Shurjo Banerjee, Wang Zhu, Jason Corso, Peter Anderson, Stefan Lee, Jesse Thomason

AI总结:

提出IVLN范式,通过持久环境中的多回合导航评估VLN智能体,发现地图构建智能体优于仅扩展隐式记忆的方法。

摘要:

我们提出了迭代视觉与语言导航(IVLN),这是一种用于评估语言引导智能体在持久环境中随时间导航的范式。现有的视觉与语言导航(VLN)基准在每个回合开始时都会清除智能体的记忆,从而测试其在没有先前信息的情况下进行冷启动导航的能力。然而,部署的机器人会在同一环境中长时间占据。IVLN范式通过训练和评估在场景游览中保持记忆的VLN智能体来解决这一差异,这些游览由多达100个有序的指令跟随的Room-to-Room(R2R)回合组成,每个回合由单独的语言指令和目标路径定义。我们提出了离散和连续的迭代Room-to-Room(IR2R)基准,每个基准包含约400个游览,分布在80个室内场景中。我们发现,扩展高性能Transformer VLN智能体的隐式记忆不足以应对IVLN,但能够构建地图的智能体可以从环境持久性中受益,这促使我们重新关注VLN中的地图构建智能体。

英文摘要:

We present Iterative Vision-and-Language Navigation (IVLN), a paradigm for evaluating language-guided agents navigating in a persistent environment over time. Existing Vision-and-Language Navigation (VLN) benchmarks erase the agent's memory at the beginning of every episode, testing the ability to perform cold-start navigation with no prior information. However, deployed robots occupy the same environment for long periods of time. The IVLN paradigm addresses this disparity by training and evaluating VLN agents that maintain memory across tours of scenes that consist of up to 100 ordered instruction-following Room-to-Room (R2R) episodes, each defined by an individual language instruction and a target path. We present discrete and continuous Iterative Room-to-Room (IR2R) benchmarks comprising about 400 tours each in 80 indoor scenes. We find that extending the implicit memory of high-performing transformer VLN agents is not sufficient for IVLN, but agents that build maps can benefit from environment persistence, motivating a renewed focus on map-building agents in VLN.

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。

↑