Hampy is a mobile food-bank assistant connected to a BracketBot. A user asks for food by voice or text, Gemini selects matching inventory, the backend dispatches one robot transfer, and the app shows live camera and movement status.
frontend/— Expo mobile appbackend/— FastAPI API, Gemini integration, inventory, robot dispatch, and camera proxyrobot/— robot-side transfer, route replay, lean-mode, audio, and live-camera programs
BBOS and its daemons are external robot dependencies and are intentionally not included. The robot programs reference the installed /home/bracketbot/bbos package and the installed ~/bbapps/quest_teleop arm utilities.
Copy robot/ to /home/bracketbot/bbapps/hampy_demo/. Preserve the recorded route at routes/table1_to_table2.json; it is replayed exactly and should only be used from the physical starting pose where it was recorded.
Start the camera service on the robot:
cd ~/bbapps/hampy_demo
uv run camera_web.py --host 127.0.0.1 --port 8082 --fps 10The backend dispatches this non-interactive workflow:
uv run transfer_both.py --execute --yesThe workflow enables lean mode for pickup, grasps with both arms, switches back to balance mode, replays routes/table1_to_table2.json, releases both items, and plays the completion audio. Keep the route and arm paths clear and keep the emergency stop ready whenever it runs.
From backend/:
cp .env.example .env
uv sync
uv run uvicorn app.main:app --host 0.0.0.0 --port 8000Set GEMINI_API_KEY, set ROBOT_ENABLED=true, and use ROBOT_TRANSPORT=ssh when the API runs on the laptop. SSH key authentication must work non-interactively for bracketbot@bracketbot-0186.local.
Forward the robot camera to the laptop in a separate terminal:
ssh -N -L 8082:127.0.0.1:8082 bracketbot@bracketbot-0186.localVerify the connections with GET /health, GET /api/robot/camera/status, and GET /api/robot/job.
Set EXPO_PUBLIC_API_URL in frontend/.env to the laptop's LAN API address, such as http://172.20.10.5:8000, then run:
cd frontend
npm ci
npx expo start --clearThe phone and laptop must be on the same network. Pressing Ask Hampy sends the request to the backend; all accepted recommendations dispatch the same two-arm table-one-to-table-two robot workflow.
cd backend && uv run pytest
cd ../frontend && npm run typecheck && npm run lint