THE ESSENTIALS IN 30 SECONDS

  • Ludo Robotics, a Krafton subsidiary, presented Ludi 0.2 at the 2026 Bay Area Robotics Symposium at UC Berkeley on October 30.
  • The company lists five areas of improvement over Ludi 0.1, including reading intent and social cues, memory, tool coordination, gesture recognition and smoother movement.
  • No benchmark scores, robot hardware or deployment timeline were published for Ludi 0.2; the only public results concern Ludi 0.1 in simulation.

Krafton, the South Korean publisher behind PUBG, has shown a new version of the AI system its robotics subsidiary is building to help machines understand the people around them. Ludo Robotics, headquartered in Palo Alto with a research lab in Seoul, presented Ludi 0.2 on October 30 at the 2026 Bay Area Robotics Symposium at UC Berkeley, according to a company press release.

What changes from Ludi 0.1

The first version, introduced in August 2026 alongside an arXiv paper, paired a vision-language model handling reasoning and dialogue with separate navigation and manipulation capabilities. It could follow spoken instructions and ask clarifying questions. Krafton says Ludi 0.2 improves on five fronts: interpreting intent and reacting to feedback and social cues, keeping memory and reasoning consistent as tasks change, coordinating navigation, manipulation and interaction tools, recognising human gestures, and moving more smoothly.

SiliconANGLE reports that the company is targeting settings ranging from warehouses and industrial sites to restaurants, kitchens and living rooms. Ludo describes its method as a mix of simulated environments, AI-generated social interactions and real-world human behaviour data, and says it is building benchmarks covering personalities, preferences, relationships and gesture-driven manipulation.

The leadership overlaps heavily with Krafton's: CH Kim is CEO of both companies, and Krafton's chief AI officer Kangwook Lee serves as Ludo's CTO. Lee said in the release that responding to people "remains a major challenge" for robots, even as their physical skills improve. Krafton links the effort to its game work, including co-playable AI characters developed with Nvidia.

What has not been shown

EconoTimes points out that no benchmark scores were released for Ludi 0.2, and that the announcement named neither the robot hardware nor a deployment date. The only published figures concern Ludi 0.1, which the August paper tested in 28 simulated scenarios in the AI2-THOR environment; a fine-tuned Qwen3.5-9B model succeeded in 20 of them, while GPT-5.6 with an optimised prompt succeeded in 26. Those results do not measure the new version or any physical robot trials.

The same outlet notes that the symposium programme listed a Ludo Robotics talk without naming Ludi 0.2. It also cites Fast Company reporting that Kim described Ludi as a research demonstration not yet reliable for everyday use, and that a Ludi 1.0 foundation model, meant to unify perception, dialogue and physical action, is planned for 2027. Krafton's own announcement gives no target date for that version.