Superb AI's second Physical AI brand film, "Right now, somewhere," shows how robots learn using the 3D space, 4D action, and intelligent object assets developed during Phase 2 of Korea's Sovereign AI Foundation Model project. The 93-second film is part of the same series as the first installment released in July. Like its predecessor, it is available in Korean, English, and Japanese on Superb AI's official YouTube channel.

TL;DR

• The film's protagonist is a virtual world replicated from reality. Within this world, robots learn how to manipulate objects through repeated trial and error.
• Every dataset figure featured in the film comes directly from the Phase 2 results: 50 3D space assets, 5,000 4D action assets, and 10,000 intelligent object images. The film also shows the selection process through which 5,000 of 7,500 recorded actions qualified as training assets.
• Like the first film, this installment tells its story entirely through on-screen text, without narration. All three language versions are being released on the same day.

Another Plate Breaks, but No One Flinches

The film opens with a door swinging open and a plate shattering. No one in the office even looks up. That's because everything is happening inside a virtual world.

The first film centered on the decision to give AI eyes. The second answers the question that follows: Where does a robot with eyes go to practice?

Our answer is a virtual environment built as a replica of real Korean homes. As the film puts it:

"Not a world someone built, but a world copied from reality."

Every Number in the Film Comes from Phase 2

The three figures that count up on screen at around 0:22 match the results documented in our Phase 2 report.

The screening sequence at 0:28 is the part we put the most effort into. Of the 7,500 actions recorded, 5,000 passed the criteria required to become training assets. Actions that failed to meet the standards for real-world scale, ground alignment, and the functionality of eight types of joints received a REJECTED stamp.

This scene captures a fundamental requirement: for a virtual world to serve as training data, it must not only look like reality but also behave like it.

Production Notes

The Robot Is the Reference Point. The World Is the Protagonist.

During planning, we considered making the robot the main character. But Superb AI's role is to operate a Physical AI Data Factory, while robots are the ones that validate its output.

That is why the camera stays focused on the place where robots learn—the replicated world. The robot appears as proof of that world's quality.

Robot Failure Is Part of the Concept

The robot fails repeatedly throughout the film. The on-screen message at 0:56 captures this idea:

"Here, it's okay to fail."

The practice counter, which climbs from 12 to 4,912, reinforces the same message.

These failures represent the safe repetition that a virtual environment makes possible. They are distinct from defects in the training assets themselves. The sequence of the film makes this distinction clear: defective data has already been filtered out during the earlier screening process.

Three Languages, No Narration

As with the first film, we relied entirely on on-screen text, with no voice-over.

Of the 44 text elements, 38 were translated into English and Japanese. The opening and closing phrase, "Right now, somewhere," follows the same construction across all three language versions to preserve the film's bookend structure. Numerical graphics retain the original figures, with only the units translated.

A Consistent Visual Language Across the Series

The opening hook begins with an everyday scene before turning it on its head with a deadpan line. Sequential text reveals, animated counters, and a two-line ending built around the progression from "Beyond X" to "Toward Y" all carry forward the visual language established in the first film.

Related Reading

• 50 Korean Homes Became Robot Training Data: Building 50 3D Spaces, 5,000 Behaviors, and 10,000 Object Assets
• Superb AI “Proprietary AI Foundation Model Project” Phase 2: Key Takeaways on Digital Twin Assetization
• Building the Eyes and Hands of Physical AI: Advancing to Phase 3 of Korea’s Sovereign AI Foundation Model Project

Frequently Asked Questions

Q. Where are the virtual spaces shown in the film based on?

They are replicas of 50 Korean home environments captured during Phase 2 of Korea's Sovereign AI Foundation Model project. The environments combine 3D Gaussian Splatting backgrounds with physics meshes for interactive objects. No information that could identify individual households is included.

Q. Are the practice counter figures (12 → 4,912) actual measured statistics?

No. The counter is a conceptual graphic illustrating how repeated training accumulates in a virtual environment.

The figures that can be cited as actual project results are 50 3D spaces, 5,000 4D actions, 10,000 object images, and the selection of 5,000 qualifying actions from 7,500 recorded demonstrations.

Q. Are the English and Japanese versions available on the same channel?

Yes. All three language versions are available separately on Superb AI's official YouTube channel. You can access them directly through the links above.

Interested in partnering with Superb AI to build training data for Physical AI? Tell us about your project below, and a Superb AI expert will get in touch.

Superb AI is a Vision Intelligence company that transforms visual data from industrial environments into actionable intelligence for enterprises. Using the 3D space, 4D action, and intelligent object assets developed through Korea's Sovereign AI Foundation Model project, Superb AI operates a synthetic data production pipeline for robot training.