摘要
Abstract
The core of embodied intelligence lies in the autonomous intelligent closed loop formed by the full-chain collabo-ration of perception,decision,execution and feedback,rather than the mere superposition of isolated technical modules.Taking this technological logic as its starting point,this paper takes companion robots as examples,systematically explores the evolutionary pathways,core bottlenecks,and design strategies of embodied intelligent robots under this paradigm.The study reveals that the intelligence level of companion robots is constrained by the weakest link in the closed loop,which comprising coarse perception leads to misguided decisions,rigid decision-making results in ineffective execution,and ex-ecution deviations without feedback correction prevent the system from continuous evolution.At the technical level,this pa-per analyzes key technological advances and challenges,including 3D semantic mapping and multimodal affective percep-tion,collaborative decision-making based on vision-language-action(VLA)models and world models,natural action genera-tion driven by end-to-end reinforcement learning,and data closed-loop feedback leveraging simulation platforms.On this basis,the concept of IP embodiment is proposed as a critical pathway for companion robots to transform from functional tools into emotional companions.Taking Disney's Olaf Robot as a representative case study,the paper examines how film and television IP can achieve large-scale deployment through technology reuse platforms.Finally,a three-tier design framework,which comprising the embodied closed-loop technology layer,IP affective empowerment layer,and scenario ex-perience layer,is established to provide theoretical reference and practical guidance for the product definition and industri-alization pathways of companion robots.关键词
具身智能/陪伴机器人/影视IP/视觉-语言-动作模型/深度强化学习Key words
Embodied Intelligence/Companion Robot/Film and Television IP/Vision-Language-Action Model/Deep Rein-forcement Learning分类
社会科学
