디자인 기반 시각 인지 재구성 -인간-로봇 시지각 비교를 통한 형태 인식 어포던스 모형 연구-

초록

Recent advances in artificial intelligence have expanded robotic systems from automated machines into intelligent agents capable of perceiving and reasoning about their surroundings. However, research in robotic vision has predominantly focused on improving the technical performance of object recognition, while overlooking the cognitive structures needed to interpret the semantic context or possible actions afforded by objects. This study reinterprets human form perception from a UX perspective and proposes a design-based form perception–affordance model that enables robotic visual systems to perform contextual understanding and autonomous action inference. Drawing on an integrated review of cognitive psychological theories—Gestalt principles, object recognition theory, and the concept of affordance—this study shows that human visual cognition infers potential actions from object forms through the interaction between bottom-up perceptual cues and top-down feedback. Based on this analysis, the perceptual mechanism is reframed as a UX-oriented experiential structure and developed into design-driven cognitive principles applicable to robotic perceptual systems. Building upon these insights, the proposed model conceptualises human cognition as a cyclical structure of perception–meaning–action. It explains how a robot perceives structural features of objects, integrates perceptual cues with contextual information through large language models (LLMs), and feeds the interpreted results back into action planning. This framework enables robots to infer the functional meaning of shapes and interpret unfamiliar environments in a manner closer to human visual perception. This study expands design from a means of visual expression to a medium for structuring cognition and experience, thereby offering a UX-based integrative perspective for robotic perception grounded in human form recognition. It highlights a promising direction for design-centred cognitive research that supports humanlike contextual understanding and action inference in robotic systems.

키워드