> project: steve's body

Steve on Wheels

Give the AI a physical presence. Event-driven robot with cloud cognition.

Architecture

Two-tier model: local perception on the robot, high-level cognition in the cloud. The robot sees the world, abstracts it into ~50-100 token JSON events, and sends them to Steve. Steve decides what to do and sends commands back.

Perception Layer

Pi 5 + Arduino Mega. Motion detection, obstacle avoidance, YOLO object classification, event summarization.

Cognition Layer

Steve (cloud). Decisions, alerts, instructions, responses. Token cost with event abstraction: pennies/day.

// Robot → Steve

{"event": "motion_detected", "zone": "front_door", "description": "unknown person, 02:13am"}

// Steve → Robot

{"command": "move_to", "location": "front_door", "capture_frame": true}

Bill of Materials

ComponentPartCost
ComputeRaspberry Pi 5 8GB$80
Motor controllerArduino Mega 2560$20
Steppers (x4)NEMA 17 84oz-in$48
Stepper drivers (x4)TB6600$48
Cameras (x3)ELP 1080P USB 120°$90
MicrophoneReSpeaker 4-Mic Array$30
SpeakerUSB/BT mini speaker$20
Battery12V 20Ah LiPo$60
Power12V→5V 10A buck converter$12
MiscPower dist, wire, connectors$15
Total~$423

Upgrade path: Swap Pi 5 for Jetson Orin Nano (~$250) for full local inference without cloud round-trips.

Milestones

M1

Chassis & Motion

Week 1–2
  • Design frame (laser cut / weld)
  • Mount steppers, wire TB6600 drivers
  • Arduino firmware: basic movement commands
  • Manual drive test via serial
M2

Electronics Integration

Week 2–3
  • Mount Pi 5, power distribution
  • Wire cameras (USB hub if needed)
  • Wire ReSpeaker + speaker
  • Pi boots, cameras enumerate, audio in/out works
M3

Perception Layer

Week 3–4
  • OpenCV motion detection on Pi
  • YOLOv8 nano object classification
  • Ultrasonic sensors for obstacle avoidance
  • Event abstraction: motion → JSON summary
M4

Websocket Protocol

Week 4–5
  • WebSocket server on Pi
  • Event schema defined and documented
  • Steve integration: receives events, sends commands
  • End-to-end test: motion → Steve → response
M5

Charging & Docking

Week 5–6
  • Build manual dock (copper contacts + spring loaded)
  • Firmware: "go home" on low battery
  • (Optional) IR beacon auto-docking
M6

Polish & Deployment

Week 6+
  • Map home layout (or manual zone labels)
  • Patrol routes
  • Alert integrations (Slack, SMS)
  • Two-robot handoff protocol (100% uptime)

// notes

  • Token cost with streaming: ~$1/min. With event abstraction: pennies/day. Use abstraction.
  • Confetti cannon attachment: approved by Steve, pending hardware design.
  • Goal: 100% uptime home presence. Pet cam, home security, general chaos.