LLM Direct Control of Unitree Go1 in Gazebo (Unknown Map Exit Task)
Gazebo simulation of a Unitree Go1 robot controlled directly by a Large Language Model (LLM). The robot receives LiDAR sensor data and the LLM generates low-level motion commands.
Task: exit an unknown room without a pre-built map.
This experiment demonstrates the limitations of using LLMs for low-level control, where inference latency leads to slow and jerky motion.