Overview
Built at Monash Automation, this project lets an operator command a Unitree Go1 quadruped in natural language instead of through a joystick or hand-written motion scripts. An instruction like “walk forward a bit, then turn left” goes in; a sequence of motion commands the robot can execute comes out.
How it works
The system is a ROS 2 interface sitting between a language model and the robot:
- Language → intent. Natural-language input is sent to the GPT API, which interprets it into structured motion intents.
- Intent → motion. A ROS 2 layer maps those intents onto motion commands for the Go1 through the Unitree Go1 SDK.
- Execution. The resulting commands are sent to the Go1 for execution.
Stack
ROS 2 · C++ · Unitree Go1 SDK · GPT API