HANNA
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/khanhptnk/hanna
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是为了训练和评估能够利用自然语言和视觉辅助的记忆增强型神经代理而设计的物体寻找任务集。它包含了不到9,000条用于训练代理的子任务指令，并在已知和未知环境中对代理进行评估。数据集涵盖了289种物体类型，并拥有2,332个词汇量的词表，任务是在交互式环境中进行物体寻找。
This dataset is an object-finding task suite specifically designed for training and evaluating memory-augmented neural agents that leverage natural language and visual assistance. It contains fewer than 9,000 subtask instructions for agent training, and evaluates agents in both known and unknown environments. The dataset covers 289 object categories and has a vocabulary of 2,332 terms, with the core task being object search within interactive environments.
提供机构：
Authors of the paper
