General strategy for autonomous drone

I’d like to know how to build a “hello world” autonomous drone prototype that I can write custom code (python would be nice) to control.

Here’s what I have so far:

  • Custom large drone that flies well by R/C, lots of extra power avail, and lots of extra space to mount things
  • Jetson TX2 development kit with latest JetPack version installed (development kit will be mounted to drone)
  • PixHawk
  • Cameras and Indoor/Outdoor GPS

After that is all connected and working, and simple ground control software like Mission Planner can fly it, what is the general strategy for writing software to fly this thing autonomously? For example, let’s say I want to fly over to a general area based on GPS. However, when I get close to the destination, I want to switch over to using custom logic based on vision inference suggestions from the Jetson and a camera. How could I approach that? It would be nice to automatically:

  1. Take off at a given time
  2. Start to fly over to a certain location with GPS
  3. Use vision to stabilize
  4. Engage a motor to perform a task
  5. Fly back home

Any suggestions would be greatly appreciated!

Hello, you may be interested in NVIDIA’s open-source Project Redtail drone…it uses deep learning inference for control and navigation.

Here’s the GitHub repo:

It’s also compatible with platforms using Pixhawk autopilot. And to mount the Jetson onboard a drone, you may be interested in trying a miniature carrier board that adopts the same credit-card sized footprint as the module itself, like some of those included from the Ecosystem List.