ChatGPT-Driven Unitree R1 Integration

Embodied AI Engineer — Unitree R1 EDU

My Unitree R1 EDU is already running on an NVIDIA Orin NX. I need an engineer to build the software bridge that allows the robot to listen, understand, see and physically act through natural voice interaction.

Target Pipeline

Microphone → STT (local Whisper preferred / OpenAI Speech API acceptable) → ChatGPT Function Calling → ROS 2 → Unitree SDK2 → Robot

Target Behaviour

* Understand commands such as “raise your hand”, “dance”, “wave”, “turn around”
* Use RGB/depth cameras to answer questions about the environment
* Return answers through voice
* Target response time of a few seconds
* At least 5 different physical actions triggered by voice

Technical Requirements

* ROS 2 + Unitree SDK2
* NVIDIA Jetson Orin NX
* Python or C++
* Modular architecture with replaceable STT and LLM layers
* OpenAI Function Calling for deterministic mapping of intents to ROS 2 actions
* Camera/VLM integration for visual questions
* Architecture allowing ChatGPT to be replaced later by a local 2B model
* Auto-start on boot
* Clean, documented and deployable code

Acceptance Criteria

1. “Wave” → R1 performs a visible waving motion.
2. “What colour is the block?” → robot analyses the camera image and provides the correct answer.
3. Low end-to-end latency, optimized for Orin NX.
4. System starts automatically after reboot with no manual terminal commands.
5. GPT can be replaced by a local model through configuration without rewriting the ROS 2 control layer.
6. At least 5 voice-triggered physical actions demonstrated on the real robot.
7. Provide source code, documentation and deployment instructions from a fresh JetPack installation.

Please apply only if you have real experience with ROS 2, Unitree SDK2, NVIDIA Jetson/Orin and LLM/VLM or Embodied AI integration.

Remote freelance work is acceptable.