Humanoid-Recognition-Guided Generative Design: Integrating Biomimetic Semantic Features and Human–AI Co-Creation for Creative Furniture Concept Development

Posture recognition provides a foundation for linking humanoid perception to the conceptual stages of human-centered furniture design. This study develops a posture-informed generative design framework that derives biomimetic semantic features from skeletal keypoints, spinal curvature, joint alignment, and anthropometric parameters. These features serve as interpretable controls within a conditional generative model capable of producing posture-responsive furniture concepts. Three modes of designer interaction — manual construction, feature-based manipulation, and semi-automated exploration — were examined in a controlled human subject study to evaluate how semantic feature accessibility and automation level influence design outcomes and learning. Designer learning was assessed using item response theory (IRT), and creative performance was evaluated through novelty, ergonomic utility, and aesthetic measures. The results show that posture-derived semantic features substantially improve designers’ ability to reason about form–function relationships, while abstract latent features provide limited cognitive support. Feature-based interaction achieved the most effective balance between exploration and ergonomic alignment. In contrast, high levels of automation increased posture-fit performance but reduced interpretability and inhibited learning. These findings indicate that humanoid-perception-guided semantics offer a meaningful structure for supporting biomimetic reasoning, enhancing the transparency of computational design tools, and enabling more informed human–system co-creation. The proposed framework extends posture understanding toward creative design applications relevant to humanoid robotics and human-centered systems.