For many people, digital twins may represent a dress to go shopping or a cup to drink tea, but the objects of digital twins are much more than that. They can even be as big as a factory or the entire planet. At GTC21 last year, Nvidia announced the creation of the digital twin of the Earth through the Modulus framework based on the physical ML model and the Omniverse platform. To create such a massive digital twin, Nvidia even built an AI supercalculator, Earth-2.
Twin by small and large numbers
The purpose of creating earth's digital twins is obvious, that is, to study meteorological science through digital twins. Last year's extreme weather had a severe impact on the global semiconductor industry, and the simulation and prediction of severe weather can effectively reduce the impact of this non-human obstruction. AI computing is already delivering hundreds of thousandfold increases in computing power, but for weather science, it's going to take millions of steps to reduce computing time, especially if we start developing deep learning predictions, which is why a supercomputer system is needed to build a digital twin earth.
Four days ahead forecast for Typhoon Mangkhut after data import/Nvidia
In addition, the earth digital twin can be further improved with the expansion of the system. Currently, it takes 16 hours of training time for the system with 128 Gpus to make the 7-day weather prediction in 0.25 seconds. However, the accuracy of the input system is not high, and it can only achieve the spatial resolution of 30km. Nvidia plans to scale it up to 16,384 Gpus and receive data with 5km accuracy, making it possible to make 7-day predictions in just 4 seconds over 200 hours of training.
This digital twin earth is interactive, can predict, visualize, monitor and track extreme weather, and can be optimized for longer forecasting cycles, not just for the weather, but for the global climate. Such digital twinning can provide some protection for countries experiencing more extreme climates, such as Japan, which has incorporated digital twinning into its "Society 5.0" strategy. In the age of AI, ignorance of natural risks is no longer an option.
Cultural digital twin
The application of digital twin is not only limited to solving current challenges, but also to preserving and protecting culture. While many museums have been closed during the pandemic, some have begun scanning physical exhibits to create twin digital collections that visitors can view from all angles without leaving their homes. This kind of digital twin has a relatively complete process, the required steps are not complex, can also be said to be one of the highest commercial degree of digital twin applications.
But digital twinning is not just a copy of something existing, it can also be used to preserve something from the past. Last year, for example, XR Content distribution company Shape Space created a digital twin before the demolition and renovation of Baishizhou, the largest urban village in Shenzhen. This digital twin application plays a very good reference role in urban reconstruction planning, while preserving the old cultural appearance of the city. Such as the old buildings, the old town and so on.
Higher challenges for 3D modeling
At present, the most difficult part of digital twin is modeling. The nvidia Omniverse mentioned above only solves the problem of physical accuracy, but these USD still need to be built by designers, creators and engineers themselves. In addition, when solving real world problems, digital twins are no longer satisfied with the simulation of virtual situations, but more concerned with the replication of real world objects. Applications such as autonomous driving and digital factories require realistic simulations of real-world scenarios, but the complexity of 3D modeling has limited the adoption of digital twinning technology in the industry because it involves recruiting professional 3D modelers, like BMW, which built digital twinning factories using Nvidia's Omniverse platform. The laborious process took 2.5 months to make.
But in order to realize the metasomes vision, non-professional users must also participate in the creation, and the way they participate in 3D modeling is through 3D scanning. The positioning of 3D scanning equipment ranges from entry-level to professional, and product forms cover mobile phones to 3D scanners. Capturing Reality, a 3D scanning App for Mobile phones, has been acquired by Epic. The App allows users to create digital twin models by taking multi-angle pictures of objects and importing them into other AR, VR and 3D content platforms.
The App is only available for iOS, though an Android version won't be available until later this year. This may be because the iPhone's camera system is more uniform, making it easier to adapt, while Android phones use sensors of varying sizes and precision. However, there are still a few defects in this 3D scanning method, such as low sensor accuracy, strict Angle and lighting conditions, mature 3D reconstruction technology, and high requirements on the number of photos taken. Finally, cloud processing is required.
In order to pursue higher accuracy, 3D scanners are still needed, but even 3D scanners are still not effective in the construction of large digital twins, such as the above mentioned reproduction of an entire village in the city, for the pursuit of efficiency, drones and other equipment. Obviously, the challenges of physical accuracy have been largely solved, but modeling still requires breakthroughs in tools and sensors.