A STUDY ON INTEGRATING DIGITAL TWIN PIPELINES USING USD & AI SIMULATION

This research study proposes the utilization of Universal Scene Description (USD) in conjunction with visual effects (VFX) and digital twin pipelines, elucidating its correlation with artificial intelligence (AI) simulation. USD is an open 3D scene technology developed by Pixar and is the de facto industry standard for sharing and editing complex 3D data across multiple digital content creation (DCC) tools and applications. In this study, the optimal methodology for integrating the VFX production pipeline with real-time digital twin simulation was explored by leveraging USD. The specific focus was on linking USD with the NVIDIA Omniverse platform to enable collaborative design and physical simulation simultaneously and to enable reinforcement learning agents to learn in a physically accurate virtual environment. Experiments were conducted using real-world complex VFX scene data to measure the performance of the proposed method. Metrics such as scene loading speed, memory usage, viewport FPS, and simulation frame rate were analyzed. The findings indicate that the USD-based pipeline exhibits superior performance in terms of data loading speed, effective memory management, and real-time simulation efficiency when compared to conventional methodologies. In addition, the feasibility of technical integration with RL frameworks such as NVIDIA Isaac Gym, RLlib and Stable Baselines3 [7] is demonstrated. Furthermore, the potential effectiveness of the proposed method for AI simulation is examined. The proposed method contributes to both the fields of visual effects (VFX) and industrial digital twins and provides a foundation for future USD-based fusion pipelines.