I think you can simulate this all in advance and have each drone executing a pre-arranged gcode-like program. I'm guessing the complexity of choreography is similar to something like autorouting a circuit board.

Other comments mention that using GPS for relative positioning is fine. The largest source of error in GPS is ionospheric delays, which will affect all these drones equally, so I can believe that. Meanwhile, the flight computer handles figuring out how much power each motor needs to move or maintain its position. What if there is high wind on the day that you want to do this? Either a single drone can remain stationary in that amount of wind, or it can't; if it can't, then you cancel the performance because there is no computer program you can write to suddenly give the drone more control authority.

I doubt there is any additional complexity here; I doubt the drones are looking out for other drones and coordinating with each other. They just go to a set position at a set time at a set speed. Any complexity lies in the positioning; whether GPS + barometric altimeter + IMS (gyro/accelerometer) is enough to get the desired accuracy. It probably is.