EdgeSculpt-HO: A hybrid optimization AI model for Real-Time 3D sculpture and pottery design

Abstract

Digital sculpting is becoming increasingly important in creative design, education, and cultural heritage preservation. Yet, existing techniques for 3D sculpture and pottery modeling often face significant limitations, including poor adaptability to user input, computational inefficiency, and inadequate responsiveness in real-time or edge-computing environments. These challenges hinder intuitive interaction and dynamic design exploration. To overcome these barriers, this work introduces EdgeSculpt-HO, a Hybrid Optimization AI model for real-time, gesture-based 3D sculpture and pottery creation. The framework integrates four key modules: SPFeat-FuseNet for extracting rich multimodal features from spatial, stylistic, and temporal data; EdgeSculptNet, a NAS-enhanced 3D generator optimized for edge deployment; C-GreyGenSelect, a chaotic Grey Wolf and Genetic Algorithm-based selector for robust feature reduction; and the Touch2Form Interaction System, which enables real-time sculptural manipulation using gestures, supported by reinforcement learning, Vision Transformers, and RNN-based haptic feedback. Notably, the system collects Internet of Things-based tactile and environmental input, and utilizes software-defined networking to dynamically manage low-latency data routing between modules across edge devices. Tested on three datasets—6K Sculptures, Art Images, and 3D Model Samples—EdgeSculpt-HO outperformed MeshGAN, AtlasNet, and 3D-StyleGAN, achieving a Dice Similarity of 0.97, Chamfer Distance of 0.0023, 94.6% optimization accuracy, System Usability Scale of 97, Mean Opinion Score of 4.6, and Net Promoter Score of + 72, validating its artistic quality, responsiveness, and deployment readiness.