Window Cleaning AI Drone MVP Development

Designed and delivered a full-stack autonomous drone system for AI-driven window cleaning — from bare PCB to flying, vision-guided hardware.

Built a custom STM32-based flight controller running BetaFlight v4.5, controlling 4 aerial + 2 cleaning motors with 30-min flight endurance. The controller streams telemetry to a Google Coral Dev Board (Linux), which runs onboard AI inference for real-time surface detection and adaptive cleaning path generation.

Simulated and validated the entire flight path planning and precision docking sequence in Gazebo, with ROS2 nodes handling navigation, motor command arbitration, and sensor fusion — before a single motor was spun on real hardware. Docking used vision-based alignment, reducing positioning error to within inches of target windows.

WiFi + BLE + RF communication stack enables a companion app for supervised control during the AI learning phase.