Memory-Maze

Memory-Maze是一个专为盲人导航设计的虚拟环境数据集，由早稻田大学创建。该数据集包含一个迷宫般的虚拟环境和来自人类记忆的新路线指令数据。数据集通过在线和现场两种方式收集自然语言指令，现场收集的指令更长且用词更多样。Memory-Maze旨在解决盲人在公共空间中导航的问题，通过模拟盲人从有视力的旁观者那里寻求路线指令的场景，提供了一个高难度的测试平台，用于开发和评估视觉语言导航模型。Memory-Maze is a virtual environment dataset specifically designed for blind navigation, created by Waseda University. This dataset includes a maze-like virtual environment and novel route instruction data sourced from human memory. Natural language instructions are collected through two modalities: online and on-site. Instructions obtained via on-site collection are longer and demonstrate greater lexical diversity. Memory-Maze aims to address the navigation challenges faced by blind individuals in public spaces. By simulating the scenario where blind people seek route instructions from sighted bystanders, it provides a high-difficulty testbed for developing and evaluating vision-language navigation models.

搜集汇总

背景与挑战

背景概述

Memory-Maze是一个由早稻田大学创建的虚拟环境数据集，专为盲人导航设计，包含迷宫环境和基于人类记忆的路线指令数据。它通过在线和现场方式收集自然语言指令，现场指令更长且用词更多样，旨在模拟盲人获取路线指令的场景，为视觉语言导航模型提供高难度测试平台。