Kickstarting the Autonomous Drone Project

The Mission

The goal is straightforward but technically heavy: Build a drone capable of autonomous object tracking using only on-board constraints (Power, Computing, Weight). No external motion capture systems, no heavy ground stations.

The Stack Dilemma

I spent the last 48 hours deciding between ArduPilot and PX4.

ArduPilot: Extremely stable, feature-rich, but the MAVROS integration felt slightly more "legacy".
PX4: Feels more native to the ROS 2 ecosystem (specially with microXRCE-DDS agent).

Verdict: Switched to DJI Tello for the initial prototyping phase. Why? It's lightweight, has a documented Python API (`djitellopy`), and allows us to focus entirely on the Computer Vision algorithms (YOLO) by offloading the flight stability to the drone's internal controller.

Processing Unit

I need a setup that allows rapid iteration.
The Strategy: Offboard Edge Computing.
Instead of running the heavy object detection on the drone, I'll stream the video feed to a laptop (or Edge Server), process the frames with YOLOv8 to find the target (Person), and send velocity commands back to the drone over Wi-Fi (UDP) to keep the target centered.

$/ expected_architecture.yaml

hardware:
  drone: "DJI Tello (Ryze Tech)"
  camera: "720p Built-in"
  
software:
  language: "Python 3.10"
  library: "djitellopy"
  ai_model: "YOLOv8 (Ultralytics)"
  task: "Person Following (Visual Servoing)"

Next Steps

Flash PX4 firmware to the Pixhawk.
Config the offboard control switch on the RC transmitter (Safety first!).
Hello World: Make the drone takeoff 1m and land via Python script.