Summary
Krafton and Seoul National University have begun a three-year research program on world models that can predict how virtual and physical environments change after an action. The collaboration will study three areas: maintaining spatial and object consistency in longer generated videos; learning human-object interaction from game-character motion data for animation and robot movement; and testing whether AI can reason about distances, passage widths and navigable space. The partners will also operate a joint graduate research center, exchanges and internships. Krafton says the work extends simulation and behavior-prediction expertise developed for games toward robotics, following its establishment of Ludo Robotics in the United States and South Korea. The company also plans to apply findings back to games.
Hana’s perspective
I find the animation component more immediately useful than the robotics headline. Long-form video consistency and believable contact between bodies and objects remain practical limits for generated scenes and game characters. A shared research program can turn those problems into measurable engineering targets rather than demo-driven claims. The harder question is transfer: game simulations provide abundant controlled data, but real rooms, objects and human movement are less predictable. If Krafton publishes its evaluation methods and shows where simulated training fails, the work could help both character animation and physical AI mature.
Are you a writer with a great story? We'd love to hear from you: APPLY HERE