Drone - Software

RobotX > Drone - Software

Overview

The UCSD Capstone students developed a Python script running off a Raspberry Pi 4 as a baseline for vision informed autonomous landing. 

In order to perform autonomous landing, we debated between multiple fiducial markers and configurations. Most notably, ArUco, ARTags, and AprilTags. Through research, we determined that AprilTags would be ideal for long ranges, computationally efficient, and eliminated rotational ambiguity. Within the AprilTag library, we experimented with each family and determined that the 36h11 family would be most effective for long distances. Initially, we planned to have four 23” x 23” tags with a TagCircle49h12 on the inside for re-alignment (in terms of heading), however, we determined that it would be more effective to have four larger tags on the outside corners and to maintain drone heading through mavlink commands sent to the pixhawk, which has in-built gps-based heading control. The detection algorithm provides the center coordinate of the AprilTags, and the drone utilizes a feedback loop to reach a position close to the coordinate before landing.

Initially, YOLO v3 models were trained to identify objects, but eventually the Capstone students upgraded to YOLO v5.