Physical AI
Robot learning is becoming a production capability
LG's new Data Factory shows that collecting, validating and improving robot data is becoming infrastructure in its own right - not merely an AI research activity.
Physical AI is often discussed as a smarter robot model. LG's new Data Factory points to a broader requirement: structured environments, representative tasks and continuous data collection are needed before adaptable robots can perform reliably outside the laboratory.
A factory for training data
LG is building a 10,000-square-metre Data Factory at its Yangjae R&D Campus in Seoul. The facility is expected to contain several hundred robots by the end of 2026, alongside training, validation and refinement areas.
Its environments include a replicated home and a simulated manufacturing space modelled on LG's Tennessee washing-machine plant. LG expects direct and synthetic or augmented training data to total 100,000 hours by year-end.
More data is not automatically better automation
Useful robot learning depends on whether the collected data represents real variation, edge cases and recovery situations. Synthetic data can broaden the range of scenarios, but its conclusions still need to be checked against physical outcomes.
The engineering question is therefore not simply how much data exists, but whether it improves dependable performance in the intended application.
The Multitech view
Manufacturers should begin with process data: the normal cycle, acceptable quality, expected variation, foreseeable exceptions and safe recovery behaviour.
AI creates value when it helps a system manage defined variation without making the application more difficult to validate, operate or maintain.
Practical takeaways
- Treat data collection and validation as part of application design.
- Capture exceptions and recovery, not only successful cycles.
- Verify synthetic scenarios against real production outcomes.
- Keep safety and deterministic control responsibilities distinct from probabilistic AI decisions.
This briefing is an independent Multitech interpretation of LG Electronics' announcement. Multitech was not involved. Facility targets and the 100,000-hour figure are company projections rather than independently verified results.