Omniverse
We have chosen NVIDIA Omniverse as a tool for problem-solving. However, this choice is not merely adopting technology, but also detecting new visions and flows within the tools and directions presented by NVIDIA, and realizing them in reality. The various components and philosophy provided by Omniverse as open source are opening the way to construct the world in a more fundamental and emergent way, beyond building traditional digital twins, including AI, humanoids, and generative simulations. With Omniverse, we are expanding our sense and practice of redrawing the world beyond just tools for problem-solving.
"OMNIVERSE goes beyond being a tool, allowing us to imagine ways to change the structure of perception and rewrite existence and the world."
Precise simulations based on NVIDIA's next-generation physics engine Newton are emerging as a key strategy for securing both precision and speed in fields requiring high-reliability simulations, such as robotics, industrial automation, medicine, and scientific visualization.
The Mega virtual logistics center, built on this foundation, provides a generative simulation environment for the entire logistics system, encompassing collaboration of multiple robots and humanoids in large spaces, sensor feedback integration, AI-based control, and logistics flow optimization experiments. This functions as a digital training ground to pre-train and improve actual operations, including synthetic data generation, motion learning, path optimization, and scenario-based testing.
In particular, the GR00T-based humanoid strategy is designed to learn and improve not only simple motion imitation but also AI-based interactions, complex task sequencing, and environmental adaptation capabilities within Omniverse. The design of GR00T, where sensor-motor-intelligence are connected as a single flow, enables new experimental methods and design strategies through deep integration with Omniverse.
All of this operates on the USD-based Omniverse framework. Omniverse is no longer just a simple tool, but a strategic platform that allows designers, engineers, researchers, and developers to collaborate in real-time in the same space-time and quickly build and iterate large-scale 3D simulation pipelines.
Who is the Subject?
"The entity handling Omniverse, reviewing multiple simulation worlds simultaneously, and presenting conclusions is no longer human but appears to be AI."
In Omniverse, as seen by POLLUX, AI is no longer just a computational engine. Now, the 'observing subject' is AI. AI senses and coordinates complex simulation worlds, performing multi-world thinking that humans cannot access. Humans are defined as beings who interpret AI's senses and judgments, give them meaning, and propose new ethics.
Multi-simulation
- 다중의 세계(멀티 시뮬레이션)
- 다양한 시간축과 조건 변화
- 복수의 에이전트와 변수들이 동시에 작동하는 공간
Complexity Management
- 실시간으로 조율하고
- 수천 개의 시나리오를 병렬 평가하고
- 조건에 따라 최적화/학습/재구성하는 건 인간의 인지 범위를 초월한 일이 되었죠.
Therefore, naturally, the entity that judges, concludes, and configures the next experiment is moving to AI, and humans no longer need to control everything or endlessly consume results to prove their existence. This is not alienation due to technology, but the recovery of humanity through technology.
In the end, even the age of AI is the age of humans.
Not all sounds are considered music. Only sounds that humans like become music. Human music is defined by humans. Humans have infinite possibilities but are ultimately beings with finite genetic constraints. Therefore, humans will create constraints. So, in the end, even the age of AI is the age of humans.
Source: Kang Kuk-jin, "What Will We Change?", Philosophic (2025)
How Have We Utilized It?
Project Time Reduction
Using Omniverse, we can reproduce actual spaces and facilities in a virtual environment and quickly simulate various layout arrangements and operational scenarios. By identifying and addressing potential problems before actual field work, we can reduce unnecessary rework and trial and error, significantly shortening the overall project time.
Facility and Robot Movement Verification and Optimization
In the Omniverse environment, we can precisely implement actual facilities virtually, efficiently proceeding from design to simulation and control optimization. In particular, we can repeatedly test and fine-tune facilities or robots in digital space to accurately pick up objects. This allows us to solve problems before actual facility operation and greatly improve equipment performance and reliability.
Cost Reduction and Productivity Improvement
In Omniverse, we can precisely simulate a robot's movement path, work efficiency, and collision possibilities to detect and solve problems before building an actual robot. In particular, deep learning and reinforcement learning-based robot control strategies can be safely learned repeatedly in a virtual environment, greatly reducing the cost of expensive physical robot testing or rework. The high-difficulty task learning of humanoid robots, which has recently been attracting attention, can also be repeatedly verified cost-effectively through Omniverse, increasing both feasibility and productivity.