Hello everyone. I want to perform a computer vision application with on-board processing on a multicopter for a university project. I have a Pixhawk 4, Pixhawk Cube, and 2.4.8. I also have a Nvidia Jetson Nano to use as a companion computer.
Which platform would be best in this scenario among Dronekit, FlytOS, ROS, or Maverick and why? Also, is it necessary to flash ApSync on the companion computer.
I have so far tried Dronekit but could not make a connection as it said Mavlink down. I am using a UART-TTL (FTDI) converter for the connection between Pixhawk and Nano.
Maverick : abandonned since some time
FlytOS : it is basically ROS with paid feature, not sure it still work with ArduPilot
ROS: unless you want to do some ROS developpement or use their tool, don’t use it.
Dronekit : maybe working, IDK.
alternative : mavproxy, pymavlink, blueos
You need to define exactly who does what ? What does the drone do ? Auto takeoff on a button, go on a target aera and start searching for somebody and stay over some finding ?
or you got an human in the loop ?
there is numerous way to develop this but you need to clarify your chain of command and what operation are done if you want to chose a tool.
Basically, you could use a camera that do detection, have a python script that do random search over a defined area and stop the drone and put it over the detection. that is a few hundreds line of code and not a need a huge framework
I will tell you a scenario of how the drone will be used.
Consider there is a drowning person at the beach and the lifeguard knows a general location of the drowning person but not exact. The lifegaurd feeds a certain area on the GCS by making a polygon and the drone automatically forms an optimal path to cover the area. It then automatically navigates the area and starts searching, if the drowning person is detected, the drone gets close to it to deploy a life-saving object such as a floating tube or anything that floats well. Once this object is detached from the drone, it means the person has received it and the drone returns to launch.
I have a decent knowledge of computer vision so that hopefully won’t be a problem. The problem is the difficulty in getting the Nano to talk to the Pixhawk.