What can a video game developer teach robots about working with people? South Korean gaming company KRAFTON, best known for PUBG: BATTLEGROUNDS, believes its experience creating interactive virtual worlds could help solve one of robotics’ persistent challenges: understanding human behaviour. Through its subsidiary Ludo Robotics, the company is developing AI that enables robots to interpret intentions, respond to feedback and adapt their actions during conversations with people.
KRAFTON has introduced Ludi 0.2, an updated robotic intelligence system designed to bring social understanding into physical AI. The technology was presented on October 30 at the 2026 Bay Area Robotics Symposium at the University of California, Berkeley.
The move into robotics builds on KRAFTON’s experience developing games in which artificial intelligence must respond to unpredictable human players. In multiplayer gaming environments, people constantly change their strategies, communicate with others and react to unexpected situations. According to KRAFTON, the AI techniques developed to handle these interactions could also help robots operate alongside people in the physical world.
From virtual teammates to physical robots
KRAFTON has invested in machine learning, reinforcement learning, large language models and multimodal AI for its gaming activities. Working with NVIDIA, the company has developed Co-Playable Characters (CPCs), AI-controlled teammates designed to understand players and respond to them during gameplay.
Ludo Robotics is now exploring how similar capabilities can be applied to physical machines.
The challenge is that working with people requires more than recognizing spoken commands. A robot operating in a workplace or home must also understand context, recognize uncertainty and adjust its behaviour when someone changes their mind.
For example, a person might ask a robot to retrieve an object, interrupt it halfway through the task and then point to a different item. Rather than continuing with the original instruction, a socially intelligent robot should recognize the correction and adapt its actions accordingly.
This is the type of interaction Ludo Robotics is attempting to develop.
“For nearly two decades, KRAFTON has built games and virtual worlds where people and intelligent systems interact in real time,” said CH Kim, CEO of KRAFTON and Ludo Robotics. “Through Ludo Robotics, we’re building on that experience to explore how AI can better understand people and interact with them naturally in the physical world.”
Combining conversation, memory and movement
The company’s first system, Ludi 0.1, was introduced in August 2026. It combined a vision-language model for reasoning and dialogue with separate capabilities for navigation and object manipulation.
The architecture enabled robots to interpret spoken instructions, maintain conversational context and ask clarifying questions when instructions were ambiguous. It also supported changes to ongoing tasks.
The underlying research was described in an August 2026 preprint, Ludi 0.1: An Agentic System for Socially Intelligent Robots. The paper explains how the system combines interactive speech, multimodal reasoning, memory and physical actions through an agent-based architecture.
Ludi 0.2 builds on that approach by expanding the system’s ability to interpret gestures, account for social cues and maintain context during changing situations. KRAFTON also reports improvements in coordinating navigation and manipulation, as well as smoother physical movement.
The objective is to bring communication and physical action closer together. Instead of treating a conversation and a physical task as separate processes, the robot should be able to use information from human interaction to guide its movements.
“Robots are becoming increasingly capable of performing physical tasks, but understanding and responding to people remains a major challenge,” said Kangwook Lee, Chief AI Officer of KRAFTON and CTO of Ludo Robotics.
The company has not disclosed comparative performance figures for Ludi 0.2 or demonstrated that the system can reliably handle unrestricted human interactions. The reported improvements therefore remain research-stage developments rather than evidence of deployment-ready social intelligence.
Using gaming experience to train social intelligence
One of the connections between gaming and robotics lies in training data.
Video games provide environments where AI systems can encounter different personalities, unpredictable decisions and changing objectives. Physical robots face similar uncertainties, although mistakes in the real world can have consequences that do not exist in a simulation.
Ludo Robotics is combining simulated environments, AI-generated social interactions and real-world human behaviour data to train its systems.
The company is also developing evaluation methods to measure how robots respond to social situations involving different preferences, personalities and relationships. These tests include interpreting gestures and using them to guide physical manipulation.
This research addresses a challenge that becomes increasingly important as robots move beyond controlled industrial environments. Traditional industrial robots generally operate within tightly defined tasks and safety boundaries. Robots intended to share everyday spaces with people must deal with instructions that are incomplete, inconsistent or expressed through gestures rather than words.
Towards a unified robot intelligence model
Ludo Robotics is already working on its next generation, Ludi 1.0. The planned foundation model aims to integrate perception, dialogue and physical action within a single AI system, rather than relying on several separately coordinated components.
The research is being conducted from the company’s headquarters in Palo Alto, California, and an additional laboratory in Seoul, South Korea. Its team includes researchers specializing in machine learning, embodied AI, robot learning and human-computer interaction.
For KRAFTON, the project is an extension of its gaming research into physical AI. The company is applying techniques developed for interactions between players and virtual characters to machines that must understand and respond to people in real environments.
