人工智能培训

搜索

人工智能论文:VALAN:视觉和语言代理导航(VALAN: Vision and Language Agent Navigation)

[复制链接]
wwwtjsix 发表于 2019-12-9 13:27:12 | 显示全部楼层 |阅读模式
wwwtjsix 2019-12-9 13:27:12 189 0 显示全部楼层
人工智能论文:VALAN:视觉和语言代理导航(VALAN: Vision and Language Agent Navigation)VALAN是一个轻量级且可扩展的软件框架,用于基于SEED RL架构的深度强化学习。该框架促进了嵌入式代理的开发和评估,以解决诸如Matterport3D和Google StreetView等逼真的环境中的基础语言理解任务,例如视觉和语言导航以及视觉和对话导航。我们在SEED RL的顶部添加了最少的抽象集,使我们能够泛化该体系结构以解决各种其他RL问题。在本文中,我们将描述VALAN的软件抽象和体系结构,并提供一个使用VALAN设计代理以进行指令式室内导航的示例。
VALAN is a lightweight and scalable software framework for deep reinforcementlearning based on the SEED RL architecture.The framework facilitates thedevelopment and evaluation of embodied agents for solving grounded languageunderstanding tasks, such as Vision-and-Language Navigation andVision-and-Dialog Navigation, in photo-realistic environments, such asMatterport3D and Google StreetView.We have added a minimal set of abstractionson top of SEED RL allowing us to generalize the architecture to solve a varietyof other RL problems.In this article, we will describe VALAN's softwareabstraction and architecture, and also present an example of using VALAN todesign agents for instruction-conditioned indoor navigation.人工智能论文:VALAN:视觉和语言代理导航(VALAN: Vision and Language Agent Navigation)
URL地址:https://arxiv.org/abs/1912.03241     ----pdf下载地址:https://arxiv.org/pdf/1912.03241    ----人工智能论文:VALAN:视觉和语言代理导航(VALAN: Vision and Language Agent Navigation)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

wwwtjsix当前离线
新手上路

查看:189 | 回复:0

快速回复 返回顶部 返回列表