Ludi 0.2 shown at UC Berkeley
Ludo Robotics, the KRAFTON subsidiary that builds AI for robots, presented Ludi 0.2, the second version of its robotic intelligence system, at the 2026 Bay Area Robotics Symposium at UC Berkeley, the company said. The company describes the system as intended to read human intent, adjust to changing surroundings and work alongside people rather than only carry out preset commands.
CH Kim, chief executive of both KRAFTON and Ludo Robotics, said the company was drawing on nearly two decades of building games and virtual worlds to study how AI can understand people and interact with them in physical settings. KRAFTON publishes PUBG: BATTLEGROUNDS.
Five areas the company lists
KRAFTON said Ludi 0.2 improves on its predecessor in five areas: interpreting human intent and social cues and responding to feedback; holding context and reasoning across changing tasks; coordinating the capabilities used for navigation, manipulation and interaction; recognising and reacting to gestures; and moving more smoothly in supported environments.
Ludi 0.1, introduced in August 2026, paired a vision-language model for reasoning and dialogue with dedicated navigation and manipulation functions, and was designed to take spoken instructions, keep conversational context and ask clarifying questions. That earlier version is set out in a paper on arXiv.
Ludi 1.0 is in development
The company said its longer-term roadmap includes Ludi 1.0, a foundation model now in development that it aims to make a single system for perception, dialogue and physical action, so that a robot can keep a shared understanding of a task with a person while communicating and acting in real time.
Kangwook Lee, KRAFTON's chief AI officer and Ludo Robotics' chief technology officer, said robots are becoming more capable at physical tasks but that understanding and responding to people remains a major challenge.
Palo Alto base and Seoul lab
Ludo Robotics is headquartered in Palo Alto with an additional research and development lab in Seoul. Its chief scientist, Rob Nowak, is an IEEE Fellow and a professor at the University of Wisconsin–Madison, and its human-robot interaction lead, Yea-Seul Kim, is a former assistant professor of computer sciences there, the company said.
The company said it trains its systems on simulated environments, AI-generated social interactions and data on real human behaviour, and is building benchmarks for social situations involving different personalities, preferences and relationships, as well as for gesture-driven physical manipulation.
