Kobe is a domain-specific language for teaching children to program reinforcement-learning robots.
Kobe/
├── lexer.py, parser.py, semantic_analyzer.py # Front-end
├── priorities.py, compiler.py # IR compilation
├── codegen.py, backend.py # Code generation + live training
├── pipeline.py # End-to-end compile entry point
├── ipc/ # Electron IPC bridges
│ ├── pipeline_ipc.py
│ └── train_ipc.py
└── ide/ # Electron + React IDE
pip install -r requirements.txt
cd ide
npm install
npm run dev
- Write a Kobe program in the editor pane.
- Diagnostics, IR, and generated TorchRL training scripts appear in the inspector.
- Adjust policy sliders — descriptions update per algorithm.
- Click Train Robot to run live simulator training (powered by Kobe's built-in PyTorch actor-critic backend).
python -c "from pipeline import run; import json; print(json.dumps(run(open('example.kobe').read()), indent=2))"