피지컬 AI의 기술동향 및 국방 활용 방안 = A Survey on Physical AI and Implementation Strategies for National Defense

다국어 초록 (Multilingual Abstract)

Recent advancements in artificial intelligence, particularly the emergence of Foundation Models, are driving a paradigm shift toward "Physical AI"—systems designed to interact autonomously with the physical world. This paper analyzes the technological trends of Physical AI and explores its strategic applications within the defense sector to overcome the limitations of traditional robotics, such as a lack of generalization and adaptability in unpredictable environments. We examine the evolution of Foundation Models, including Large Language Models (LLMs) and Vision-Language-Action (VLA) models, which integrate perception, reasoning, and actuation into End-to-End embodied systems. Furthermore, this study investigates critical enabling technologies for Physical AI, categorizing them into policy learning, End-to-End architectures, value function learning, LLM-based task planning, and open-vocabulary capabilities. Lastly, based on this technological analysis, we propose four core application strategies for national defense.