检索规则说明:AND代表“并且”;OR代表“或者”;NOT代表“不包含”;(注意必须大写,运算符两边需空一格)
检 索 范 例 :范例一: (K=图书馆学 OR K=情报学) AND A=范并思 范例二:J=计算机应用与软件 AND (U=C++ OR U=Basic) NOT M=Visual
作 者:刘馨竹 王亚珅 石晓军 陈思 LIU Xinzhu;WANG Yashen;SHI Xiaojun;CHEN Si(Artificial Intelligence Institute of CETC,Beijing 100043,China)
机构地区:[1]中国电子科技集团公司信息科学研究院,北京100043
出 处:《无人系统技术》2025年第2期108-122,共15页Unmanned Systems Technology
基 金:国家自然科学基金(U22B2061)。
摘 要:具身智能是智能体通过与环境交互进行感知、决策、行动的能力,是人工智能的前沿研究领域。对2024年具身智能的相关研究和发展动向进行了综合评述,并对未来发展方向进行展望。首先分析了具身智能基础模型的技术研究进展;随后讨论了环境认知理解技术的发展动向;接着分析了任务行动执行技术的最新进展;最后概述了具身智能未来的重要发展方向。综述表明,具身智能基础模型正在探索扩展性和通用性更强的三维场景表示方式、“视觉-语言-动作”框架以及三维场景重建技术;三维识别、三维图谱构建以及三维空间理解推理等环境认知理解技术,聚焦形成泛化能力更强以及可适应未知开放场景的认知能力;任务行动执行技术关注更精准的动作执行策略以及更高效灵活的协同策略;未来具身智能模型将向着具备更精准三维空间感知认知能力、更加通用、轻量化、灵活鲁棒、易扩展、易部署的方向发展。Embodied intelligence refers to the capability of agents to perceive,make decision and act through interaction with the environment,and it is a frontier research field in artificial intelligence.This paper overviews the related research and development trends of embodied intelligent technology in 2024 and envisions the development direction in the future.Firstly,the technical progress of the basic model of embodied intelligence is analyzed.Then,the technology development trends of environment perception and understanding are discussed.Subsequently,the latest development of task action execution strategies is analyzed.Finally,the important development direction of embodied intelligence in the future is summarized.The survey shows that:(i)The basic models of embodied intelligence are exploring more extensible and more general 3D scene representation method,visual-language-action framework and 3D scene reconstruction technology;(ii)The scene recognition and understanding technology including the 3D recognition,3D graph construction,and 3D spatial understand focuses on constructing the cognition ability with stronger generalization ability and adaptability to unknown open-world scenes;(iii)The task action execution technology focuses on more accurate action strategies as well as more efficient and flexible collaboration strategies;(iv)In the future,the embodied intelligence model will be developed in the direction of more accurate 3D spatial perception and cognition,more general,lightweight,flexible and robust,easy to extend as well as easy to deploy.
关 键 词:具身智能 人工智能 环境交互 基础模型 环境认知理解 任务行动执行
分 类 号:TP183[自动化与计算机技术—控制理论与控制工程]
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在载入数据...
正在链接到云南高校图书馆文献保障联盟下载...
云南高校图书馆联盟文献共享服务平台 版权所有©
您的IP:216.73.216.222