Yuntao Wang, Xiaolin Niu, Jianle Ba, Zhou Su, Linkang Du
Driven by recent advances of large models and agents, embodied artificial intelligence (AI) emerges as a transformative paradigm for next-generation AI, endowing agents with physical forms and the ability to perceive, reason, and act within real-world environments. Unlike disembodied or virtual AI agent systems, embodied agents co-evolve cognition, control, and embodiment through continuous feedback loops, enabling applications ranging from humanoid robots to autonomous vehicles. In this survey, we first introduce a dual-brain architecture of embodied AI and examine its foundational technologies and key characteristics. We then analyze the security and privacy landscape, identifying critical vulnerabilities and evaluating existing/potential countermeasures. Finally, we outline emerging trends and open research directions in this emerging field, charting a roadmap toward efficient, secure, and ethically aligned embodied AI ecosystems.