Semi-autonomous Obstacle Avoidance for UAVs

Pixhawk
1

Hello everyone,

I’m working on my final-year project and I’d like to ask for your advice. My goal is to build a quadcopter with semi-autonomous obstacle avoidance. The drone will normally be flown manually by the pilot, but when the Raspberry Pi detects an obstacle nearby (using camera and sensors), it should temporarily take control, send a setpoint to the Pixhawk to avoid the obstacle, and then return control back to the pilot.

The title of my project is: “Development of a drone with obstacle avoidance using camera, sensors, and artificial intelligence.”

Hardware I plan to use:

  • Pixhawk 2.4.8 flight controller

  • Raspberry Pi 5 as companion computer

  • Raspberry Pi Camera Module v2

  • LiDAR / ToF sensor for additional obstacle detection

My main goal is just to deliver a simple demo within ~6 months — for example, the drone being able to detect and avoid a single obstacle (like a wall) during flight, not a full-scale autonomous navigation system.

My questions:

  1. Is this approach (Pi5 + Pixhawk 2.4.8 + YOLO for detection + MAVLink setpoints for avoidance) realistically feasible within that timeframe?
2

I can’t say if it’s feasible, that depends on many factors, certainly possible in theory.

What I can say is that I wouldn’t use a Pixhawk 2.4.8 to avoid any problems with knockoff hardware that might or might not work properly. I recommend using any of the supported or manufacturer supported hardware as listed in Flight Controller (Autopilot) Hardware | PX4 Guide (main) .

In terms of sending setpoints to the RPi5, you have some choices:

  1. Setpoints via MAVLink, either directly using MAVLink or a library like MAVSDK.
  2. Setpoints via ROS 2, using the ROS 2 bridge.
4

Thanks for your input! I completely understand your point about the Pixhawk 2.4.8 not being ideal for production or long-term reliability.

This is only for my university thesis project (“Development of a drone with obstacle avoidance using camera, sensors, and artificial intelligence”). The goal is just a simple demo: the Raspberry Pi detects an obstacle and sends a short avoidance command (e.g., move sideways) to the flight controller, then gives control back to the pilot.

So it doesn’t need to be a fully robust system just enough for one or two avoidance maneuvers during a test flight. Do you think even with a Pixhawk 2.4.8 this is still feasible for such a small demo, or would it definitely fail?

5

If it doesn’t work properly, then that’s unfortunately your answer.