In the Programming course the Matternet drone is shown landing on top of a Mercedes-Benz without an Aruco marker. How was this possible, and how does one approach the challenge of precision landing without a marker (eg, a distant yet known object)? Could AI provide a solution? For example, you want to land on top of a structure and you might have some landmarks on the roof you could use to guide the vehicle, yet no pre-determined marker is placed?
Good catch! The Matternet drone landing on the Benz I believe was using the IR-lock system, as opposed to the computer vision based method we use in the course.
I bet AI could be used as a solution, as long as it could spit out some x and y distance errors from the intended land position to the drone. Then you could package those same error coordinates into angle form, and send the same precision landing mavlink message that we use with the pure opencv method.
As a side note, I’ve tried in the past to PL with only GUIDED velocity commands. This turns on to be a horrible method, probably because the latency is too high and you aren’t capitalizing on the PL code in the AC firmware that beautifully handles the PL re-positioning given the input error coordinates.