Open Grounded Planning
收藏资源简介:
Open Grounded Planning数据集是由中国科学院软件研究所构建,旨在评估大型语言模型在开放域环境中的规划能力。该数据集涵盖日常生活、工具使用和机器人控制等多个领域,包含超过76,000个任务,每个任务都有详细的任务目标、约束条件和候选行动集。数据集的创建过程涉及从WikiHow等来源收集原始数据,并将其转换为统一的格式。该数据集主要用于研究如何使AI系统在开放和可执行的行动集中生成可执行的计划,解决AI在复杂任务规划中的挑战。
The Open Grounded Planning Dataset was developed by the Institute of Software, Chinese Academy of Sciences, aiming to evaluate the planning capabilities of large language models in open-domain environments. This dataset covers multiple domains such as daily life, tool use, and robot control, and contains more than 76,000 tasks, each with detailed task objectives, constraints, and candidate action sets. The dataset creation process involves collecting raw data from sources including WikiHow and converting it into a unified format. This dataset is primarily used to research how to enable AI systems to generate executable plans from open and executable action sets, addressing the challenges faced by AI in complex task planning.




