Earticle

초록

영어

This study analyzes the characteristics of China's Embodied AI (Physical AI, 具身智能) policy and legal system with a focus on the humanoid robot industry, and derives regulatory innovation tasks for Korea. As AI evolves beyond generative AI toward Embodied AI that interacts directly with the physical world, China has been among the most active policy drivers. The Ministry of Industry and Information Technology issued a national-level humanoid-specific policy document, the Guiding Opinions on the Innovative Development of Humanoid Robots, in October 2023, followed by the first mention of Embodied Intelligence in the 2025 Government Work Report and the State Council's Opinions on Deepening the AI+ Initiative in August 2025. The findings are threefold. First, China has built a multi-layered governance system of policy documents first, standards following, and legislation deferred within a short period. Within roughly two to three years, a comprehensive structure spanning top-level design, sector-specific policies, standardization organizations, and local implementation has been established. Yet humanoid-specific rules rely on normative policy documents and recommended standards with weak binding force. Unlike China's general industrial promotion regimes, hard norms on physical safety, including market-entry rules and mandatory standards, are deferred, and physical AI is regulated far more lightly than content-related digital AI. Moreover, although the State Council's 2026 legislative plan calls for accelerating comprehensive AI legislation, no dedicated AI law had been enacted as of September 2026. Second, China has institutionalized a demonstration-data-standards feedback cycle through four instruments: pilot programs and scenario opening, national-local jointly established innovation centers, prudent-and-inclusive supervision combined with Chinese-style regulatory sandbox mechanisms, and science and technology ethics review. Although a Western-style sandbox is not codified in general law, the trial-and-error mechanism originating from the 2019 fintech pilot has spread to the 2023 intelligent connected vehicle pilot and the prudent-and-inclusive supervision in the 2025 Beijing Autonomous Vehicles Regulations, while the AI+ Opinions call for a trial-and-error tolerance mechanism and sandbox supervision is being discussed as a flexible regulatory tool. Innovation centers such as Beijing's Tiangong and Shanghai's Qinglong supply open-source platforms, pilot production infrastructure, and training data as public goods, and Chinese firms accounted for six of the top ten humanoid shipment vendors in 2025 according to Omdia's estimates. Third, gaps in physical safety and ethics norms, ambiguity in personal information rules for robot-collected data, and low legal predictability are structural limits. The national safety standard is being drafted as a recommended (GB/T) rather than a mandatory standard, and the applicability of the Personal Information Protection Law and Data Security Law to robot-collected data is underspecified, exemplifying the asymmetry of promotion-oriented legislation. Based on this analysis, three tasks are derived for Korea. First, legal alignment: on the foundation of the Intelligent Robot Act and the AI Framework Act, in force since January 2026 and amended effective July 2026, definitions, safety certification, and demonstration exceptions specific to humanoid robots should be established, and the gap between the two statutes closed, since the AI Framework Act already defines high-impact AI by risks to life and physical safety but confines it to enumerated sectors, so that coverage of humanoid robots used in manufacturing or households remains unclear. Second, a humanoid-specific sandbox track and demonstration platform: the Industrial Convergence Regulatory Sandbox, which approved A-ROBOT's humanoid demonstration in September 2025, should be extended to a fast-track for controlled environments where data feed back into standard-setting, and the K-Humanoid Alliance launched in April 2025 could evolve into a physical demonstration hub akin to China's joint innovation centers. Third, proactive standards and data governance: the Korean Agency for Technology and Standards should draw a roadmap for humanoid general safety and human-robot interaction standards, engage in ISO/IEC forums, and update the mobile visual device provisions of the Personal Information Protection Act to reflect always-on data collection. Combining statutory stability with China's demonstration-centered flexibility, Korea can build a differentiated, trust-based regulatory model for humanoid robots.

한국어

본 연구는 휴머노이드 로봇 산업을 중심으로 중국 Embodied AI(具身智能, 피지컬 AI) 정책⋅법제도의 특징을 분석하고 한국의 규제혁신 과제를 도출하기 위한 목적으로 수행되었다. 중국사례를 통해 우리나라에 다음과 같은 시사점을 제공한다. 1) 지능형로봇법⋅AI기본법 기반의 휴머노이드 정의⋅안전인증⋅실증특례 정합화, 2) 산업융합규제샌드박스의 휴머노이드 특화 재설계와 실증-데이터-표준 환류 체계 구축, 3) 휴머노 이드 안전⋅데이터 표준의 선제적 정비와 국제표준 대응 관련 내용이다. 본 연구를 통해 분석한 연구결과는 다음과 같다. 첫째, 중국은 2023년 <휴머노이드 로봇 혁신발전지도의견>부터 2025년 국무원 <인공지능+행동 심화 의견>까지 정책문건 선행-표준후속-법률 유보의 다층 거버넌스를 단기간에 구축하였으나, 휴머노이드 로봇에 특화된 규율은 규범성문건과 추천성 표준에 의존하여 구속력이 약하며, 일반 산업육성 법제와 달리 물리적 안전에 관한 경성규범의 유보와 디지털 AI⋅물리적 AI 간 규제 비대칭이라는 특수성을 보인다. 둘째, 중국은 시범사업⋅시나리오 개방, 국가⋅지방 공동구축 혁신센터, 포용적⋅신중한 감독과 중국식 규제샌드박스 기제, 과학기술윤리심사라는 제도수단을 통해 실증 - 데이터 - 표준의 선순환을 제도화하였다. 중국은 서구식 규제샌드박스를 일반법으로 제도화하지는 않았으나 핀테크 혁신 감독 시범사업, 자율주행 시장진입 시범사업, 베이징 자율주행 조례의 포용적⋅신중한 감독으로 시행착오 허용 기제를 확산시켰고, 국무원 <인공지능+ 의견>은 시행착오 허용 관리제도의 완비를 명시하였다. 셋째, 물리적 안전⋅윤리 규범의 공백과 로봇 수집 데이터에 대한 개인정보 규율의 불명확성은 촉진형 법제의 구조적 한계로 확인되었다. 이러한 분석에 기초하여 본 연구는 한국의 규제혁신 과제로 지능형로봇법⋅AI기본법의 법률 기반 위에 휴머노이드 정의⋅안전인증⋅실증특례를 정비하고 AI기본법의 고영향 인공지능 영역 열거 방식과 로봇 안전법제 사이의 공백을 해소하는 정합화, 산업융합 규제샌드박스의 휴머노이드 특화 재설계, 그리고 안전⋅데이터 표준의 선제적 정비와 국제표준 대응을 제시하였다. 향후 이 연구는 한국의 휴머노이드 로봇 관련 정책 마련과 규제 개선에 참고할 만한 시사점을 제공할 수 있을 것으로 기대한다.

목차

Ⅰ. 서론
Ⅱ. 선행연구 검토 및 분석틀
Ⅲ. 중국 Embodied AI 관련 주요 정책⋅법제 분석
Ⅳ. 휴머노이드 로봇 산업 관리체계 분석
Ⅴ. 중국 Embodied AI 법제도의 성과와 한계
Ⅵ. 결론
참고문헌
국문초록
Abstract

키워드

Embodied AI 피지컬 AI 휴머노이드 로봇 중국 법제도 규제샌드박스 규제혁신 Embodied AI Physical AI Humanoid Robot China's Legal System Regulatory Sandbox Regulatory Innovation 具身智能 物理人工智能 人形机器人 中国法律制度 监管沙盒 监管创新

저자

• 최해옥 [ HaeOk Choi | 과학기술정책연구원 혁신성장실 연구위원 ]

참고문헌

발행기관

간행물