Autonomous Projectile-Evasion Drone
Designed and built a small autonomous drone in simulation that can detect incoming projectiles and execute evasive maneuvers in real time, combining depth sensing, trajectory prediction, and automated flight control.
- Developed an automated test suite with 796 unit tests and ran 95+ live projectile trials, hitting 91.7% detection recall across the full test library.
- Split the system into ROS 2 packages for flight control, perception, and an independent guard alarm, with the dodge triggering within 150 ms of a detection.
- Dodged 78 of 78 hit-intent throws at up to 8 m/s with zero false dodges on clear-miss controls.
- Traced the speed ceiling to sensor reach and frame rate rather than maneuvering, and scoped a global-shutter camera upgrade for 20 m/s projectiles.
- Built safety in from the start: a 10 m geofence, 5 m ceiling, and a dedicated kill switch for every test.
- ROS 2
- Gazebo
- Python
- ArduPilot
- MAVLink
- Kalman Filtering
- Computer Vision