摘要
随着具身智能在服务机器人、自动驾驶等关键场景的应用扩大,其安全挑战已从算法鲁棒性延伸至行为可控性、交互社会契合性及跨文化伦理适应性。分析了具身智能在感知—决策—执行闭环中的风险传播机制,探讨了策略建模、异常检测、行为可解释性及价值对齐等关键防御路径,并指出跨模态攻击、平台异构与意识形态适配等实际部署难题。提出具身智能安全体系需从“技术可控”向“系统协同”与“社会对齐”转变,构建兼具鲁棒性、适应性和文化敏感度的综合体系,是实现可信部署的关键所在。
As embodied AI systems are increasingly deployed in critical domains such as service robotics and autonomous driving,its safety challenges have evolved beyond algorithmic robustness to encompass behavioral controllability,social compliance in interactions,and cross-cultural ethical adaptability.It analyzes risk propagation mechanisms within the perception—decision—action loop of embodied intelligence,examines key mitigation strategies including policy robustness modeling,anomaly detection,behavioral interpretability,and value alignment.It further identifies practical deployment challenges such as crossmodal attacks,platform heterogeneity,and ideological adaptation.It proposes that the security paradigm for embodied AI must shift from"technical controllability"toward"system coordination"and"social alignment",emphasizing that constructing integrated frameworks with robustness,adaptability,and cultural sensitivity constitutes the cornerstone of trustworthy deployment.
作者
贾雅清
徐雷
陶冶
王芃
Jia Yaqing;Xu Lei;Tao Ye;Wang Peng(China Unicom Research Institute,Beijing 100048,China;China United Network Com-munications Group Co.,Ltd.,Beijing 100033,China)
出处
《邮电设计技术》
2025年第7期41-45,共5页
Designing Techniques of Posts and Telecommunications
关键词
具身智能
系统安全
多模态鲁棒性
策略可解释
Embodied intelligence
System safety
Multimodal robustness
Policy interpretability
