Krafton's Ludo Robotics Unveils Ludi 0.2 AI to Help Humanoid Robots Understand People

PUBG Publisher Krafton Advances Into Physical AI With Ludi 0.2

Krafton, the South Korean video game publisher behind PUBG: Battlegrounds, unveiled Ludi 0.2 at the 2026 Bay Area Robotics Symposium at UC Berkeley1

1

4

4

. Developed by Krafton's Palo Alto-based subsidiary Ludo Robotics, the robot intelligence model represents a significant step toward creating socially intelligent AI that can interpret human intent and work alongside people in physical environments2

2

. The presentation was led by Kangwook Lee, Krafton's Chief AI Officer and CTO of Ludo Robotics, who explained that while robots are becoming increasingly capable of performing physical tasks, understanding and responding to people remains a major challenge3

3

.

Translating Gaming AI Expertise to Humanoid Robotics

Krafton believes nearly two decades of developing AI for video games provides a competitive advantage in humanoid robotics1

1

. CH Kim, CEO of both Krafton and Ludo Robotics, draws parallels between creating virtual worlds where software responds to unpredictable player decisions and building AI for robots that must adapt to human behavior. The company has been developing AI in the form of nonplayer characters even before the recent wave of interest in the sector1

1

. This work includes PUBG Ally, developed in collaboration with Nvidia, described as the world's first voice-enabled co-playable character that cooperates with players, adapts to gameplay, and remembers previous interactions1

1

. Kim stated that the foundational technology applied to PUBG Ally is very similar to the technology being applied to Ludi, with the longer-term goal of expanding this work into developing proprietary foundation models1

1

.

Enhanced Capabilities in Ludi 0.2 for Natural Human-Robot Interaction

Building on Ludi 0.1 introduced in August 2026, Ludi 0.2 advances several capabilities that support more natural human-robot interaction3

3

. The system now features improved gesture recognition, allowing robots to interpret human gestures and physical movements alongside verbal commands2

2

4

4

. Enhanced conversational memory enables the system to maintain context during changing tasks and respond to feedback more effectively1

1

4

4

. For example, a robot carrying a requested drink can stop and switch to another beverage when a person changes the instruction mid-task4

4

. The system also demonstrates improved coordination between movement and object manipulation, along with smoother physical actions4

4

.

Technical Architecture Combining Multiple AI Models

Ludi 0.1 uses an architecture combining several existing AI models to achieve its capabilities1

1

. A vision-language model from Alibaba's open-source Qwen family analyzes camera images and transcribed speech to decide what the robot should do next1

1

. Nvidia's GR00T robotics foundation model, fine-tuned by Ludo, controls physical tasks such as picking up objects1

1

. The system runs on Unitree's G1, a roughly $16,000 humanoid robot, with Ludo's software coordinating the components1

1

. The company trained its reasoning model on 14,432 computer-generated interaction sequences, while the robot learned physical manipulation separately from recordings of people remotely controlling it in a mock household1

1

.

Leveraging Synthetic Data From Game Engines

Krafton believes its gaming expertise offers advantages in generating training data for AI for robots1

1

. Game engines simulate how objects move and interact, providing virtual environments that can generate synthetic data1

1

. Kim says Ludo is already using synthetic data from these worlds and plans to incorporate more real-world information as its research advances1

1

. However, while simulation makes it easy for physical AI to prepare robots to work with objects and around humans, it doesn't always prepare them to model human behavior2

2

. This gap highlights why interpreting human intent and responding to social cues from people around the robot remains elusive, even as startups bring increasing capability to help robots see objects and manipulate them2

2

.

Research Roadmap Toward Unified Foundation Models

Ludo Robotics is developing Ludi 1.0, a next-generation foundation model currently in development that aims to integrate perception, dialogue, and physical action within a unified system3

3

4

4

. The objective is to integrate communication, reasoning, and physical action into one brain that can watch what's happening, listen to commands and speech, interpret those commands alongside human gestures, and reason about the environment in real time2

2

. This would coordinate similarly to how another human decides on actions based on another person's actions in their workspace2

2

. Kim acknowledges that everyone seems to be working on building a foundation model, but considers it an enormous challenge1

1

. Kim's longer-term goal is to expand this work into developing Krafton's own proprietary foundation model1

1

.

Establishing Silicon Valley Research Presence

Ludo Robotics is headquartered in Palo Alto, California, with an additional research and development lab in Seoul, South Korea3

3

4

4

. These locations place the team close to leading research talent, academic institutions, and the broader technology ecosystem3

3

. Krafton announced that Ludo Robotics has officially opened its headquarters research center in Palo Alto, which will work alongside its Seoul office to advance AI and robotics research4

4

. The company's scientific leadership brings together expertise across AI, robotics, and human-computer interaction3

3

. Chief Scientist Rob Nowak is an IEEE Fellow and professor at the University of Wisconsin-Madison, specializing in machine learning, embodied AI, and robot learning3

3

. Yea-Seul Kim, Human-Robot Interaction Lead, is a former assistant professor of Computer Sciences at the University of Wisconsin-Madison, with expertise in human-computer interaction, information visualization, and data science3

3

.

Market Context and Competition in Humanoid Robotics

The robotics industry is pushing toward making humanoids part of everyday life, taking on work in homes and offices1

1

. Billions of dollars are flowing toward machines capable of increasingly complex physical tasks1

1

. The humanoid robot maker Figure raised more than $1 billion last year at a $39 billion valuation, while the robotics software startup Skild AI secured $1.4 billion in January to develop artificial intelligence that can control different kinds of robots1

1

. However, working alongside people requires understanding incomplete instructions and adapting when someone changes their mind midway through a task1

1

. A robot that needs constant explanations could leave its owner supervising the very work they hoped to hand over1

1

. Kim sees initial opportunities in homes, offices, and factories, where robots must work alongside people rather than simply perform repetitive tasks1

1

. This is the Holy Grail of physical AI, where intelligent robots can perform actions not just in industrial environments and warehouses, but also retail and domestic spaces such as restaurants, living rooms, and kitchens2

2

.