ECONOMY

S. Korean pre-startup team nurtured by Carbon Black finishes 2nd at Nvidia AI competition

by
Shin Dong-yoon
Published : Nov. 19, 2026 - 21:49:26
    • Copy Completed!

View Korean Original

Team wins industry innovation award at Nvidia's first V2D challenge

Technology converts human task footage into robot training data

Object position error kept within 1 centimeter

Choi Jae-gu, leader of DDD, a pre-startup team nurtured by Carbon Black [Carbon Black]
Choi Jae-gu, leader of DDD, a pre-startup team nurtured by Carbon Black [Carbon Black]

DDD, a pre-startup team nurtured by Carbon Black — a venture studio specializing in physical AI — has placed second in its category at an international robotics AI competition while also winning an industry innovation award.

According to Carbon Black, DDD finished second in Track 3 of the Video to Data (V2D) Challenge hosted by Nvidia and received the Industry Innovation Award, demonstrating its technical competitiveness on the global stage ahead of its formal launch.

The V2D Challenge, held for the first time this year, is an international competition that evaluates technology for converting footage of humans performing tasks into training data for robots. The contest ran for about a month and a half, and the results were announced Friday at CoRL 2026, the Conference on Robot Learning.

The central challenge of the competition was extracting the information needed to train robots on their movements from ordinary video footage, without a separate, complex data-collection process. For a robot to carry out tasks in a real-world environment, it requires training data that captures the shape, position and movement of objects.

DDD showcased technology that reconstructs the structure and movement of objects appearing in video in three dimensions and converts that reconstruction into simulation data for robot training. Particularly notable was its precision in restoring the position of objects in footage to within roughly 1 centimeter of error.

The Industry Innovation Award is given separately from competition rankings, based on an assessment of a technology's originality, economic viability, level of automation and potential for use in real industrial settings. DDD was recognized for the degree of automation in its technical implementation and the potential to reduce data-generation costs. Judges included Changliu Liu, a professor at Carnegie Mellon University; Wenjie Li, a Columbia University professor and head of robotics at World Labs; and Javier Romero of Meta.

Nvidia's competition organizers and judges described DDD's submission as an innovative, highly automated approach, and singled out its visual rendering quality as among the best of the Track 3 entries.

Choi Jae-gu led the technology's development. Choi entered and graduated at the top of his class from Seoul Science High School before earning both a bachelor's and doctoral degree in electrical and computer engineering from Seoul National University. His research has focused on recognizing the structure and movement of objects in three-dimensional space, and he has published findings in IEEE Robotics and Automation Letters and at the International Conference on Intelligent Robots and Systems.

"I am glad to have had the practical value and precision of the 3D recognition technology I have been developing recognized on the world stage," Choi said. "If robots can acquire training data from ordinary video footage alone — without expensive dedicated equipment — it could help cut the time and cost required for data collection."


realbighead@heraldcorp.com
This content was produced with the assistance of AI translation services.

MOST READ