arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2210.08246cs.HC
收藏

一种用于在与机器人交互时理解推理过程的用户界面

A User Interface for Sense-making of the Reasoning Process while Interacting with Robots

Chao Wang, Joerg Deigmoeller

AI总结:

针对传统知识图谱可视化工具无法结合机器人具身特性辅助专家理解推理系统的问题,研究开发了支持自然语言指令发送、推理过程可视化、3D实例关联和第一人称执行展示的交互界面,并通过实际使用收集了设计参考反馈。

摘要:

由于知识图谱有潜力弥合常识知识与移动机器人平台可执行能力推理之间的差距,近年来将知识图谱整合到机器人系统中受到了越来越多的关注。此前,图谱可视化已被开发者广泛用于理解知识表示。然而,由于缺乏现实世界环境的抽象知识与机器人动作之间的关联,传统的可视化工具并不适合专家用户结合机器人的具身特性来理解、测试、监督和修改基于图谱的推理系统。为此,我们开发了一个界面,使机器人专家能够用自然语言向机器人发送指令,随后该界面会可视化机器人将指令映射为函数以在常识知识库中查询的过程,将结果与3D地图中的现实世界实例相关联,并从机器人的第一人称视角展示机器人的执行过程。机器人专家在日常开发中使用该系统3周后,我们收集到了一些反馈,这为此类系统的设计提供了洞见。

英文摘要:

As knowledge graph has the potential to bridge the gap between commonsense knowledge and reasoning over actionable capabilities of mobile robotic platforms, incorporating knowledge graph into robotic system attracted increasing attention in recent years. Previously, graph visualization has been used wildly by developers to make sense of knowledge representations. However, due to lacking the link between abstract knowledge of the real-world environment and the robot's actions, transitional visualization tools are incompatible for expert-user to understand, test, supervise and modify the graph-based reasoning system with the embodiment of the robots. Therefore, we developed an interface which enables robotic experts to send commands to the robot in natural language, then interface visualizes the procedures of the robot mapping the command to the functions for querying in the commonsense knowledge database, links the result to the real world instances in a 3D map and demonstrate the execution of the robot from the first-person perspective of the robot. After 3 weeks of usage of the system by robotic experts in their daily development, some feedback was collected, which provides insight for designing such systems.

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。

↑