Back home

Ludi 0.2

x 0.80
Visit source

Intro

Ludi 0.2 is the second version of the robotic intelligence system from Ludo Robotics, a Palo Alto-based subsidiary of KRAFTON — the Korean game company behind PUBG: BATTLEGROUNDS. It was presented on October 30, 2026 at the Bay Area Robotics Symposium (BARS) at UC Berkeley.

Ludi is built for socially intelligent robots: rather than only executing fixed commands, it aims to understand what a person actually wants, keep track of the conversation, read gestures and adapt as the task changes.

Explain

Architecture (from Ludi 0.1 paper and launch materials)

  • Reasoning core: a fine-tuned vision-language model (Ludi 0.1 used a Qwen3.5 VLM) trained on multi-turn human–robot interaction traces — ambiguous requests, clarifications, corrections, interruptions.
  • Harness: a purpose-built agent loop that decides which "tool" to call at each step: speak, navigate, or manipulate.
  • Skills: specialized navigation and learned manipulation policies; the demo runs on a Unitree G1 humanoid with a Dex-3 hand and fine-tuned NVIDIA GR00T manipulation policies.

What is new in 0.2

  • Social intelligence – better interpretation of intent, feedback and social cues.
  • Memory & reasoning – keeps context through changing tasks.
  • Tool coordination – smoother combination of navigation, manipulation and dialogue.
  • Gesture interpretation – recognizes and responds to human gestures.
  • Physical movement – more adaptable motion.

Roadmap

Ludo says it is working toward Ludi 1.0, a unified foundation model integrating perception, dialogue and physical action. Critics noted the 0.2 announcement came without public benchmarks.

Usage

How to follow it

Who should care

  • Robotics researchers working on human-robot interaction and agentic VLA systems.
  • Teams building service or home robots that must handle vague, spoken instructions.
  • Game-AI watchers: Ludi builds on KRAFTON's Co-Playable Character work with NVIDIA.

There is no public SDK or weights for Ludi 0.2 at launch; it is a research system demo.