Speaker:	Ryuki Hyodo (SpaceData Inc. / Institute of Science Tokyo)
Title:	A Concept for a Three-Layer AI-Driven Digital Twin Platform
Date (JST):	Fri, Apr 18, 2025, 13:30 - 14:30
Place:	Lecture Hall
Abstract:	In recent years, digital twins have garnered considerable attention in a wide range of fields, including semiconductor manufacturing, healthcare, logistics, autonomous driving, and even urban development. A digital twin is a technology platform that recreates real-world entities as “twins” in a virtual space, sometimes enabling real-time or near-real-time analysis and simulation. This presentation introduces our research group’s original digital twin platform developed using Unreal Engine. Our platform is composed of the following three layers: • Layer 1 (Reproduction of Physical Space): This layer reproduces the geometry of structures and terrains—such as the International Space Station, cities, or the lunar surface—establishing fundamental physical and geographical boundary conditions. • Layer 2 (Reproduction of Environmental Parameters): This layer standardizes various environmental condition data, such as temperature, humidity, and radiation, making it possible to run complex environmental simulations. • Layer 3 (Implementation of Arbitrary Physical Laws): This layer integrates arbitrary physical laws, such as air resistance and electromagnetic phenomena, in a format that can be computed on GPUs or HPC (High-Performance Computing) environments. Building on these three layers, the platform employs AI to simulate human, object, and agent decision-making processes in the digital twin environment with high fidelity. In practice, methods such as reinforcement learning and agent-based modeling are combined to support tasks like exploring optimal solutions for urban development, disaster prevention scenarios, and crowd mitigation strategies. Our platform is also particularly valuable for academic institutions such as IPMU, offering an innovative way to visualize, store, and share research data. We think this can represent a new paradigm in data visualization and outreach. In this presentation, we will detail the technical background of the three-layer structure, our unified approach to handling environmental data and physical laws, and several use cases of AI-agent-based decision-making simulations. Notably, we will discuss the high level of extensibility and flexibility achieved by this platform, as well as the benefits of integrating data assimilation and multi-physics capabilities into a comprehensive simulation framework. Finally, we will offer perspectives on the future development and real-world implementation of digital twin technology.
Remarks:	This seminar marks the first in a joint seminar series between Kavli IPMU and F-REI (The Fukushima Institute for Research, Education, and Innovation), held under the framework of our MOU. Dr. Hyodo is the Chief Science Officer at SpaceData Inc. and is leading several exciting projects based on his "AI-Driven Digital Twin Platform.” Their approach may offer new ways to visualize, store, and share research data at IPMU.

← Back