My 14 year old is pretty good at coding and has developed a couple of nice neural networks (python) and is beginning to explore the idea of developing a drone with computer vision and gps enabled capabilities. Idea is to feed gps coordinates via a supporting software from laptop and the drone flies to the location (say across the neighborhood ) and avoid the obstacles and then land to a marked location by image recognition (gets close to it via gps but the final descent and landing spot is identified by a specific landing image/ a red cross or some thing of the nature). He does want to learn the system inside out to improve his coding and advanced nerual network skills (I think he is ivy league bound )
I am just helping out in initial research and I have landed at px4 as the route to move forward.
Now the main questions, what kind of hardware setup should we be starting out with.
Looking for an easy build for the drone frame (a kit preferred for hardware)
pixhawk 4 or cube is the way to go ? (assuming both of us are not great at soldering so looking for more of a standard cabling connections)
From what I have read to enable computer vision we would need an off board computer, should it be a Raspberry pi or some other option is recommended?
My research tells me I need a drone kit, pixhawk borad (type pending), raspberry pi or i7 or some other offboard computer, couple of stereo camera kits, distance sensors, telemetry modules?, GPS modules…
Please excuse my ignorance here as I am a newbie and have tried to read up as much as I can before going ahead and buying the items
@aabhutta good luck with your project, really cool idea! So we flying avoidance (https://github.com/PX4/avoidance) on a couple different frames. For the bigger ones we have the Intel NUC as a companion computer, for the smaller drones the Intel UP core is so far the best working option. We use Pixhawk 4 on some, on the others the Pixracer. Where I would say for you Pixhawk 4 is probably the better option, as it does not require soldering. For the frame itself I think you are quite free in choosing what suits you in size, I don’t think it matters too much for the avoidance performance, it just needs to fit your companion board. So far the best 3D cameras we used are the Occipital Structure Core or the Realsense D435.
Yes thats that’s where I landed when I started the learning process on the basics but thought needed a few steps before that having never explored this roadmap before
Great info thank you!. One point of clarification, I assume the companion board is also attached to the frame along with pixhawk 4 (then communicates with ground via telemetry)
or is the companion bird attached to the ground controller (physically usb?). Is the nvidia also a good option in terms of support and ease of configuration or intel nuc is a better approach to start out and expand down the road.
The companion computer will be attached to the frame and communicate to the flight controller over a telem port. I once set up an avoidance bench setup with the nvidia jetson TX2. So I’m pretty sure it would work as well, but we never flew that, so there might be issues not yet uncovered. In the end it is your choice which companion board you feel comfortable with. I can only vouch for the UP core and the NUC to work for sure, as we are currently using those.
Very helpful indeed, will narrow it down between NUC and UP core once I get closer to the buying and setup process. We are thinking of first setting up the components with Pixhawk 4 and a quad / hex copter kit with gps to make it work for autonomous flight. Then we will start setting up the obstacle avoidance with an off board system along with stereo cameras and TM. Once again thanks for the clear response to newbie questions. Will post more along the way once we get rolling.
)