ArraMon

ArraMon (PokeMon + Arrangement) 是一个新的任务/数据集，用于在自然语言指令的指导下学习导航+组装。 ArraMon 任务中的代理应开发综合能力，将指令中的方向短语映射到大型现实户外环境和网格布局环境（均为 3D 以自我为中心）中的动作，并了解对象引用和空间/深度表达式以执行任务成功地。一个数据集（英语；也扩展到印地语），包含人工编写的导航和组装说明，以及相应的地面实况轨迹。ArraMon (PokeMon + Arrangement) is a novel task and dataset focused on learning navigation and assembly under the guidance of natural language instructions. Agents tasked with completing the ArraMon challenge must acquire comprehensive capabilities: mapping directional phrases within instructions to actions across both large-scale realistic outdoor environments and grid-based layout environments (both 3D egocentric), as well as understanding object references and spatial/depth expressions to successfully fulfill assigned tasks. This dataset, initially available in English and further extended to Hindi, contains manually authored navigation and assembly instructions paired with corresponding ground-truth trajectories.

搜集汇总

背景与挑战

背景概述

ArraMon是一个用于导航与组装联合任务的数据集，包含人工编写的自然语言指令及对应的真实轨迹，支持英语和印地语。该数据集发布于2020年，由北卡罗来纳大学教堂山分校提供，旨在帮助代理在3D环境中学习执行导航和组装操作。