Scholarlink

Embodied Artificial Intelligence (AI)Embodied Artificial Intelligence isEmbodied AI a paradigm toward Artificial General Intelligence (AGI), bridging cyberspace and physical systems. This chapter reviews embodied AI from foundations to advances, including history, key technologies, components, architectures, applications, simulators, datasets, and evaluation metrics. Recent progress in Large Language Models (LLMs) and world models has greatly advanced this field. LLMs enhance embodied AI through semantic reasoning and task decomposition, enabling high-level instructions and natural language actions. World models provide internal representations and predictive capacities, supporting interactions consistent with physical laws. This chapter introduces LLM/MLLM-based and world model-based approaches. It explains their roles in embodied cognition and interaction. It also advocates a joint architecture that combines MLLMs and world models to enable complex task execution in physical environments.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Embodied AI

  • 章节

  • 发表日期：

    2026年

查看全文
  • Xin Wang,
  • Tongtong Feng,
  • Huaping Liu,
  • Wenwu Zhu

摘要

Embodied Artificial Intelligence (AI)Embodied Artificial Intelligence isEmbodied AI a paradigm toward Artificial General Intelligence (AGI), bridging cyberspace and physical systems. This chapter reviews embodied AI from foundations to advances, including history, key technologies, components, architectures, applications, simulators, datasets, and evaluation metrics. Recent progress in Large Language Models (LLMs) and world models has greatly advanced this field. LLMs enhance embodied AI through semantic reasoning and task decomposition, enabling high-level instructions and natural language actions. World models provide internal representations and predictive capacities, supporting interactions consistent with physical laws. This chapter introduces LLM/MLLM-based and world model-based approaches. It explains their roles in embodied cognition and interaction. It also advocates a joint architecture that combines MLLMs and world models to enable complex task execution in physical environments.