NaVQA

该数据集名为NaVQA，是一个针对长视野导航视频的问题回答数据集，其中包含了针对机器人长视野导航视频的空间、时间和描述性问题注释。该数据集基于CODa机器人导航数据集构建，提供了关注空间理解、物体检测、标志阅读、动态事件理解以及上下文推理等多种类型的问题。该数据集总共涵盖了7个序列的210个问题，其任务是针对机器人导航的长视野视频问题回答。

This dataset, named NaVQA, is a question answering (QA) dataset targeting long-horizon navigation videos for robot applications. It includes annotations of spatial, temporal, and descriptive questions focused on robot long-horizon navigation videos. Developed based on the CODa robot navigation dataset, this dataset provides questions spanning multiple categories including spatial comprehension, object detection, sign reading, dynamic event understanding, and contextual reasoning. In total, it encompasses 210 questions across 7 sequences, with the core task being long-horizon video question answering for robot navigation.

搜集汇总

背景与挑战

背景概述

NaVQA是一个用于机器人导航的长期时空记忆问答数据集，包含210个空间、时间和描述性问题，时间范围长达20分钟，旨在评估ReMEmbR系统的长时程推理能力。