How aastik0303/fruit_fly_autodrive works
NeuroPath AI combines FlyWire connectome ingestion and neuron exploration with an embodied dodge-game demonstration. The explorer workflow runs from the user through the React page and API client to backend routes and connectome tables, then displays neuron data in a 3D scene. The game workflow encodes the falling-block scene as eye input, propagates it through the connectome, and selects actions with a learned Q readout. Edges follow excerpts, SOURCE INDEX references, or README documentation; unsampled route details are not expanded.
Main entry points: game.py Game WebSocket · FlyGamePage.jsx Game interface · train.py Agent training · brain.py Connectome response
Components 20
Connectome data
- Data ingestion
backend/preprocessing/ingest.py - Processed tables
backend/preprocessing/loaders.py - Table loaders
backend/preprocessing/loaders.py
Backend runtime
- API app
backend/api/main.py - Neuron API
backend/api/store.py - Game WebSocket
backend/api/game.py - Eye encoding
backend/embodied/eye.py - Connectome response
backend/embodied/brain.py - Dodge environment
backend/embodied/dodge_env.py - Q-learning agent
backend/embodied/agent.py - Agent training
backend/embodied/train.py
Connectome explorer
- Frontend app
frontend/src/App.jsx - Neuron explorer
frontend/src/pages/ExplorerPage.jsx - REST client
frontend/src/services/api.js - Connectome scene
frontend/src/three/BrainScene.jsx
Embodied game
- Game interface
frontend/src/pages/FlyGamePage.jsx - Game socket client
frontend/src/services/gameSocket.js - Game arena
frontend/src/three/GameArena.jsx - Brain activity view
frontend/src/three/BrainActivity.jsx
Other components
- Researcher
Connections 29
- Researcher → Frontend app: opens
frontend/src/App.jsx - Frontend app → Neuron explorer: renders
frontend/src/App.jsx - Frontend app → Game interface: renders
frontend/src/App.jsx - Data ingestion → Processed tables: writes
backend/preprocessing/ingest.py - Table loaders → Processed tables: reads
backend/preprocessing/loaders.py - API app → Neuron API: registers
backend/api/main.py - API app → Game WebSocket: registers
backend/api/main.py - Neuron API → Table loaders: loads
backend/api/store.py - Neuron explorer → REST client: requests data
frontend/src/pages/ExplorerPage.jsx - REST client → Neuron API: requests neuron data
README.md - Neuron explorer → Connectome scene: renders
frontend/src/pages/ExplorerPage.jsx - Game interface → Game socket client: connects
frontend/src/pages/FlyGamePage.jsx - Game socket client → Game WebSocket: exchanges commands
backend/api/game.py - Game interface → REST client: requests point cloud
frontend/src/pages/FlyGamePage.jsx - REST client → Neuron API: requests visualization
README.md - Game interface → Game arena: renders
frontend/src/pages/FlyGamePage.jsx - Game interface → Brain activity view: renders
frontend/src/pages/FlyGamePage.jsx - Game WebSocket → Dodge environment: advances game
backend/api/game.py - Game WebSocket → Eye encoding: encodes view
backend/api/game.py - Game WebSocket → Connectome response: computes activity
backend/api/game.py - Game WebSocket → Q-learning agent: selects action
backend/api/game.py - Connectome response → Eye encoding: uses input matrix
backend/embodied/brain.py - Connectome response → Table loaders: loads connectome
backend/embodied/brain.py - Eye encoding → Dodge environment: uses game geometry
backend/embodied/eye.py - Agent training → Connectome response: trains against
backend/embodied/train.py - Agent training → Q-learning agent: trains
backend/embodied/train.py - Agent training → Dodge environment: simulates episodes
backend/embodied/train.py - Agent training → Eye encoding: featurizes views
backend/embodied/train.py - Game arena → Brain activity view: displays alongside