The Virtual World Left by the Metaverse Welcomes Robot 'Natives'

Author: Zen, PANews

On September 8, an American startup named Antioch completed a $32 million Series A funding round, led by the well-known Silicon Valley venture capital firm Greylock. With an additional $8.5 million raised in a seed round earlier this year, the company, which has been established for just over a year, has accumulated a total funding of $40.5 million.

Antioch's main approach is to transfer real-world robots, sensors, software systems, and operating environments into computers, allowing for large-scale training, testing, and validation of machines in a digital world that closely resembles reality. Failures and extreme scenarios that are costly, difficult to replicate, or even dangerous in the real world can be repeatedly simulated here.

Antioch is not an isolated case; as Physical AI becomes a new investment direction in the AI industry, a number of companies focusing on 3D spaces, digital twins, physical simulations, and synthetic data are regaining capital attention. What they are trying to solve is how to create a world in computers that is sufficiently close to reality.

Just looking at the descriptions of "building virtual worlds and replicating real scenarios" easily reminds one of the previous wave of technological enthusiasm for the "metaverse." However, the difference is that Antioch is not creating a virtual world for users to explore with VR headsets; this time, it is the robots that truly need to "live" in the virtual world for the long term.

The virtual world has not disappeared; it is just that this time, the "residents" have changed.

Humans Did Not Move into the Metaverse

In August 2021, at the SIGGRAPH conference, the most important computer graphics conference globally, NVIDIA announced an expansion of its Omniverse platform. At that time, NVIDIA was very clear about its positioning: Omniverse is a platform for simulation and 3D collaboration, and it is "providing the foundation for the metaverse."

Its core idea is to bring 3D content that is originally scattered across different software into the same virtual world.

One very important underlying technology is called USD, or Universal Scene Description. This is a 3D scene description standard originally developed by Pixar for complex animated films and later open-sourced. Simply put, it is like a "universal language" for the 3D world: what objects are in a virtual scene, where they are, what materials they use, how they move, and what relationships they have with each other can all be organized through USD.

Therefore, designers using 3D modeling software like Blender, artists using Adobe Substance 3D to create textures and materials, and engineers using different engineering software can connect their respective content to Omniverse through USD and collaborate in the same 3D world.

At that time, NVIDIA even described this future as a kind of "3D Internet": today, the internet connects web pages, and in the future, it may connect countless 3D spaces that can be entered and interacted with in real-time.

In October of the same year, Facebook changed its name to Meta, marking the beginning of the most fervent phase of the metaverse.

Compared to NVIDIA, Zuckerberg's vision of the future was even grander. In his vision, people would no longer just use the internet through a two-dimensional screen, but would enter a digital world with spatial and "presence" sensations, where they could hold meetings, work, entertain, shop, and socialize. Meta estimated that in about the next decade, the metaverse could cover 1 billion people and form a massive digital economy.

The metaverse is a basket that can hold anything. In the following two years, VR, AR, digital humans, virtual offices, 3D games, digital twins, and digital assets like NFTs were all included in a unified imagination of a "large-scale digital world."

However, the consumer-grade metaverse industry has yet to materialize, and many participants have quietly exited. Meta has not given up on VR, AR, and smart glasses, but the Reality Labs that support these businesses remain an extremely costly long-term investment. In 2025, Reality Labs' annual revenue was approximately $2.207 billion, while its operating loss reached $19.193 billion. Meta also expects that in 2026, the operating loss of this department will largely remain at the same level as in 2025.

In contrast, the changes in Omniverse are more intriguing. Five years ago, NVIDIA referred to Omniverse as "the foundation of the metaverse." Now, the definition provided on NVIDIA's official website has changed to "a set of libraries and microservices for developing Physical AI applications."

The upper narrative and product focus of Omniverse have shifted to industrial digital twins, robot simulation, and autonomous driving development. It now connects a series of tools for robots and Physical AI, such as Isaac Sim, Isaac Lab, Cosmos, PhysX, and Warp. The largest narrative that the same platform carries has changed.

In fact, the history of robot simulation, industrial digital twins, and physics engines predates the "metaverse's first year" by a long time. The previous wave of the metaverse merely placed relatively scattered technologies like 3D modeling, real-time rendering, digital twins, virtual collaboration, open 3D standards, and NFTs into a unified imagination of a "large-scale digital world."

Therefore, after the retreat of the metaverse, these infrastructures did not disappear. People have gradually discovered that robots may need a virtual world more than humans do.

Why Robots Need a Virtual World

At this year's GTC conference, NVIDIA explained the humanoid robot training system and once again showcased a very intuitive "robot data pyramid."

The bottom layer consists of the largest quantity of internet and human data, such as web pages, images, videos, and human behavior records. They are almost inexhaustible and can tell robots what the world is like and what humans are doing, but they usually lack the joint angles, forces, tactile feedback, and control signals that robots truly need to execute actions.

At the top of the pyramid is real robot data. Allowing robots to truly grasp, transport, and walk through remote operation or autonomous operation can yield data that is closest to the final task, but it is also the most expensive: a robot can only truly experience a maximum of 24 hours in a day, and it also involves machines, venues, operators, and equipment maintenance.

In between the two is simulation and synthetic data. It is not as accurate as real robot data, but it can be rapidly replicated and generated in parallel on GPUs; it does not merely "see what humans are doing" like internet videos, but can directly produce the actions, states, and sensor data required for robots to execute tasks.

NVIDIA thus proposed that it hopes to continuously expand the synthetic data layer in the pyramid in the future, making it an important data source for robot training.

The pros and cons of the three types of data are actually very simple: internet and human data are cheap and massive in scale, but they are far from what robots actually execute; real robot data is closest to actual deployment and has the highest physical authenticity, but it is expensive and slow; simulation data is in the middle—while it loses some authenticity, its scale, cost, and controllability far exceed that of real machines.

For this reason, simulation is becoming one of the directions most favored by capital and major companies in the U.S. Physical AI infrastructure.

Moreover, it solves not only the issue of data quantity. The biggest characteristic of the real world is that it cannot be copied at will. If a company has 100 robots, it can only run 100 robots simultaneously; a warehouse cannot possibly rearrange shelves, lighting, flooring, and products every ten minutes just to train robots. But in a virtual environment, the same robot can be copied into thousands or even tens of thousands of digital avatars.

A robotic arm can simultaneously face 1,000 different desktop layouts; robots can continuously change the positions, weights, and materials of objects; lighting, camera positions, ground friction, and even sensor errors can be artificially randomized. This training method is known as Domain Randomization, where robots do not need to memorize a perfect virtual environment but are intentionally exposed to constantly changing environments, allowing the model to learn to complete the same task under various changes.

Additionally, it is worth mentioning that besides needing data for "normal operation," developers also need to pay attention to data under extreme conditions. A sudden camera failure, a slippery surface under the robot's feet, or a robot falling are all scenarios that need to be considered and overcome. The real world cannot create accidents every day just to collect data, but the virtual world can.

This is also part of the reason why some believe that robot development cannot always rely on real testing grounds. Real-world testing requires equipment, venues, and engineering time, and extreme failures are difficult to replicate. But in simulation, a system update can immediately face thousands of conditions for parallel testing.

The biggest question the metaverse once faced was why humans should enter the virtual world. Now, robots have provided a completely different answer. For humans, the virtual world is merely an optional alternative to real life. But for robots, the virtual world can serve as a training ground before entering reality.

The "Virtual World" That Has Become a Business Again

A new "virtual world infrastructure" industry is emerging around Physical AI, and the divisions of labor among these companies are becoming increasingly refined.

In August 2026, NavVis, headquartered in Munich, Germany, completed a $85 million funding round.

NavVis is a spatial data company. Its core business is not training robots but quickly scanning real factories, buildings, and infrastructure into high-precision 3D data using mobile scanning devices. The company disclosed that in just 2025, its system processed and distributed over 1 billion square meters of real space.

NavVis has now directly positioned this business as the "data foundation" for Physical AI—if robots are to work in real factories in the future, the first step is to let computers accurately know what the factory looks like.

The South Korean startup NdotLight addresses another layer of issues. NdotLight is a Physical AI data company specializing in creating 3D assets for robot simulation. In August of this year, it secured 15 billion KRW, approximately $10.6 million, in new funding, led by the Korea Development Bank (KDB).

Ordinary 3D models usually only need to be sufficiently realistic to be used in games or animations. However, for robot training, this is far from enough, as it also needs to know the weight of the chair, where collisions might occur, the surface friction, and which parts can move.

NdotLight's TRINIX converts ordinary 3D content into so-called SimReady assets, which are 3D assets that can directly enter robot simulators. In addition to appearance, it also supplements mass, friction, joint structure, and collision data, and connects to NVIDIA Omniverse and Isaac Sim. It addresses how to ensure that virtual objects not only replicate reality but can also be used by robots like real objects.

On September 17, another company from Iceland, Treble, secured $18 million in funding. Treble is an acoustic simulation company. After establishing a virtual space, it can simulate how sound propagates in rooms, buildings, and other environments, generating synthetic acoustic data.

As robots, smart glasses, and other Physical AI devices increasingly rely on microphones to understand the real environment, companies are also beginning to apply their acoustic digital twin technology to Physical AI training, primarily addressing how robots can "hear" sounds that are close to reality in the virtual world.

These types of companies combined have actually begun to reassemble a digital world, which is now very close to what people referred to as the "virtual world" five years ago. The difference is that while the metaverse pursues immersion, Physical AI pursues authenticity.

This leads to one of the most important concepts in the field of robot simulation: "Sim-to-Real," which explores whether what robots learn in the virtual world can still be applied when they return to the real world.

This is also why NVIDIA's Isaac Sim emphasizes not only visual rendering but also includes rigid body dynamics, joints, collisions, and various virtual sensors to minimize this gap as much as possible; Isaac Lab further allows developers to run a large number of robot environments simultaneously, training robot strategies in parallel through GPUs.

Thus, the core difference between today's robot virtual world and many consumer-grade metaverses lies in the fact that the former's value must ultimately return to the real world for validation.

Humans ultimately did not move into the metaverse as initially envisioned. However, the roads paved for the virtual world have not been completely abandoned; robots are walking back to reality along them.