autonomous drone

Faculty: Professor Marco Levorato

Project Director : Ian A. Harshbarger

Team: Yushi Lin, Jiajun Li, Purav Patel, Eric Qiu, Tianhao Wu

[Project Github]

First, we familiarized ourselves with the working environment, including but not limited to Github repositories for UAVs, dronekit documentation, and Jetson Nano.

Then, configured VL53L0X time of flight sensor, connected it with jetson nano, and adapted Python scripts to print distance data.

Next, we built a target using cardboards and PVC pipes; collected, cleaned, and labeled a dataset of the target on Roboflow for training a YOLOv5 model.

Validated the YOLOv5 model, successfully detected the target and the stop sign with high accuracy (over 90%). We further exploited this high accuracy by dynamically calculating the percentage of the target in images captured by the onboard webcam as a safety guarantee.

Then, scripted a simulation using dronekit-silt, fixed bugs in our code, and optimized the algorithm. After trials and errors, we ran the simulation with Mission Planner and passed the test. Finally, we tested our drone in an open field.