Feeling Machine

How it works

1. 1
   Connect & Launch

   Connect to your Reachy Mini, navigate to "Applications", find "Feeling Machine" and click "Launch" (make sure you have the dashboard installed).
2. 2
   Talk naturally

   Speak to the robot like you would to a friend. Tell jokes, share good or bad news, ask questions, or just say hello. The robot won't talk back - it only expresses emotions.
3. 3
   Watch it react

   The AI interprets what you say and responds with expressive body language and procedurally-generated sounds. Each reaction is unique!
4. 4
   Try classic emotions

   Say "play classic emotion" or "show me classic joy" to use pre-recorded animations from the Pollen library instead of procedural ones.

Installation (local)

1. 1
   Clone & install

   git clone https://huggingface.co/spaces/RemiFabre/feeling_machine
cd feeling_machine && uv sync
2. 2
   Run the app

   cd src/feeling_machine && python main.py --gradio
   Then open http://127.0.0.1:7860/ or http://127.0.0.1:7861/

The PAD Model

Instead of playing pre-recorded animations, Feeling Machine generates unique expressive movements on-the-fly using the PAD emotional model - a psychological framework that represents emotions in a 3D space.

Why procedural generation?

Pre-recorded emotions look the same every time. Procedural generation creates infinite variations - each expression is unique while staying emotionally consistent.

The PAD values control head movement amplitude, speed, antenna positions, and sound characteristics. Joy produces quick, bouncy movements with high-pitched sounds. Sadness creates slow, drooping motions with lower tones.

Named emotions available

joy, happiness, anger, fear, sadness, surprise, boredom, uncertainty, disgust, neutral

Or the AI can specify custom P/A/D values for nuanced emotional expressions.

Credits

PAD-based motion and sound generation by Anaelle Jaffré, from the I3R student project.