Ludo Robotics presents Ludi 0.2, aiming to help robots read human intent

Robots running Ludi 0.2 are intended to better follow what people mean as instructions, gestures and circumstances change. Krafton said its subsidiary Ludo Robotics presented the updated robotic intelligence system at UC Berkeley’s Bay Area Robotics Symposium on October 30, 2026, with claimed improvements spanning human interaction, memory and physical movement.

Following people, not just their first instruction

Ludo’s stated focus is on robots that can understand context and adjust as an interaction develops, rather than only execute predefined commands. Krafton describes the update as progress toward combining communication, reasoning and physical action. The capabilities it lists cover both understanding a person and coordinating the machinery needed to respond.

• Human intent and feedback: Krafton says Ludi 0.2 improves its interpretation of what people want, its response to corrections and its handling of social cues.
• Memory and reasoning: The system is intended to retain context and reason through tasks and situations as they change during an interaction.
• Coordination: The update targets more coordinated use of navigation, object manipulation and interaction capabilities, rather than treating those abilities as isolated parts.
• Gestures and movement: Krafton also claims better recognition of human gestures and smoother, more adaptable physical movement within the environments the system supports.

These are company-described improvements, not quantified performance results. Krafton’s announcement provides no numerical comparison with the previous version and does not specify public access or pricing. The presentation therefore introduces an updated research system, without establishing how much more reliably it handles those interactions.

Robots are becoming increasingly capable of performing physical tasks, but understanding and responding to people remains a major challenge.

Kangwook Lee, Krafton chief AI officer and Ludo Robotics CTO

The architecture behind the social ambition

The earlier Ludi 0.1 research paper, submitted on August 22, 2026, explains the foundation for this work. That system combined interactive speech, memory and reasoning across different kinds of input with navigation and learned object-handling skills. Its decision-making core was a fine-tuned vision-language model—a model that works with visual information and language—rather than a single model directly responsible for every physical skill.

The researchers trained that core on records of interactions unfolding over multiple conversational turns. Those records included ambiguous requests, clarifications, corrections and interruptions, alongside multistep tasks and conversations mixing social exchanges with task instructions. This makes the training objective more specific than recognizing an object or responding to one spoken command: it includes revising behavior while the exchange is still underway.

A software framework managed the loop between the model and its tools, while specialized navigation and manipulation policies carried out physical skills. The paper also described a second purpose for the system: generating records combining different kinds of interaction data to help develop a more deeply integrated robot foundation model.

Training for ambiguity, with integration still ahead

Ludo’s research approach combines simulated environments, AI-generated social interactions and real-world human behavior data. Krafton identifies ambiguity, changing preferences, nonverbal communication and unpredictable behavior as difficulties in training robots for human interaction. The mix is intended to address that social complexity, not simply provide more examples of objects to move or places to navigate.

The company is also developing evaluation benchmarks for complex social situations involving different personalities, preferences and relationships. Gesture-driven physical manipulation is another evaluation target. That work gives a more concrete description of what Ludo wants to measure: how a robot handles the person and the evolving interaction, alongside its ability to complete a physical action.

Ludi 1.0 remains a longer-term project in development. Krafton says that foundation model aims to unite perception, dialogue and physical action within one system. Its goal is for robots and people to maintain a shared understanding of tasks while communicating and acting in real time—not a capability established by the Ludi 0.2 announcement.