Article

LLM agent drives an AMR using Husky A300 + ROS2 Nav2 + local Qwen3

A developer on Reddit demonstrated driving a Husky A300 mobile robot with a local Qwen3 LLM agent through ROS2 Nav2: it uses an asynchronous background task architecture, so chat still responds during navigation, and llama.cpp was compiled with CUDA 12.4 for real-time local inference with no cloud dependency. Spatial coordinates are kept in the conversation context with no external database; it currently runs entirely in simulation, and the same action-client/function-calling architecture can be transferred directly to a physical robot.

Part of the event: Developer demonstrates local Qwen3 LLM agent driving Husky A300 mobile robot

This page provides a summary and source link only.