Mythos

Inspiration

I've always been fascinated by how stories shape the way we learn and remember. From cave paintings to cinema, storytelling has been our most powerful teaching tool. I wanted to see if a robot usually built for utility, could instead become a medium for expression. Mythos grew from that idea: using AI and robotics not to build or assemble, but to perform and tell stories in a new, unexpected way.

What it does

Mythos connects GPT-OSS with a LeRobot SO101 arm to create an AI storyteller that makes physical gestures:

• Voice input triggers story generation with synchronized robot movements
• AI generates scripts, Manim visualizations, and coordinated physical actions
• Robot performs gestures (talk, listen, think, nod) timed with generated audio
• Projection mapping creates immersive visual experiences

Why GPT-OSS?

GPT-OSS excels at four critical capabilities that make this possible:

1. Structured outputs - Reliable JSON plans for consistent tool integration
2. Step-by-step reasoning - Coordinating robot actions, visuals, and audio timing
3. Code generation - Quality Manim visualizations created on-demand
4. Tool calling - Direct robot control through function calls

How I built it

Architecture:

• MCP Server: Python server exposing robot actions as LLM tools
• Backend: FastAPI handling STT (Whisper), TTS (Kokoro), Manim rendering
• Frontend: React grid interface for projection
• Hardware: SO101 robotic arm via lerobot SDK

Pipeline: Voice → Transcription → GPT-OSS Planning → Synchronized TTS + Robot + Video

What's next

• YOLO + SAM2 integration for intelligent object highlighting
• Educational applications for interactive teaching

Built With

• gpt-oss
• sam2
• whisper