XYZ Trains Robots on Data From 50 Live Service Sites
Korean robotics firm XYZ says it trains robots on data from 50 sites run by its subsidiary, then validates them in Nvidia-powered simulation before deployment.

Korean robotics company XYZ has detailed how it develops physical AI: by teaching robots with data gathered at real service sites. CEO Hwang Seong-jae presented the approach during the physical AI startup session at Nvidia AI Day Seoul 2026, held at COEX in Seoul on November 9 and 10.
Physical AI refers to AI that moves and works directly in the physical world, as robots do. Hwang argued that a good AI model alone is not enough for robots to make their own decisions in real environments. Field data collection, training in virtual environments and execution on actual robots, he said, must be linked in a single development pipeline.
A loop between field data and simulation
XYZ collects data from 50 sites operated by a subsidiary. It uses Nvidia's robot simulation technology to train and validate robots in virtual environments, then applies the results to real robot tasks. The company says the aim is a cycle in which field data and virtual training strengthen each other, so robots can handle a wide range of work settings.
The company is focusing on robots that go beyond understanding a person's request to planning and carrying out long tasks on their own. Its in-house robot intelligence framework, BrainX, connects customer interaction, task planning and physical manipulation. XYZ is applying it to robots such as DEUX, a semi-humanoid with an upper-body-focused design.
Folding clothes in half the time
XYZ also presented a research result: a robot that folds clothes cut the time needed from 50 seconds to 25 seconds while reaching a 96% success rate. The company describes it as a case where not only accuracy but also the speed required in real services was improved.
At the same event, XYZ took part in the AI Startup Partnership Forum hosted jointly by South Korea's Ministry of Science and ICT and Nvidia. Twenty-two domestic startups working in AI models, physical AI, infrastructure and semiconductors attended.
Apps worth a look
Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.
Talk to an AI conversation partner; every sentence comes back corrected and explained in your language. 9 languages, A1 to C2.
Open EDIFACT and X12 messages, see every segment explained, convert to JSON/XML, export PDF and Excel. Runs locally, no upload.
Pick who it is for, the occasion and their personality. The search engine reads product descriptions from local shops and ranks what fits. Free, no sign-up.
More from this day
See all →Nvidia and Unity Join Emulate's $700m Fundraising Round

Seoul Labs launches LetsTicket, a blockchain ticketing app

South Korea Orders Official Probe Into GS E&C Security Breach


