View Technologies can track a moving RAIN Rfid tag to the size of a softball in a xyz plane in realtime.
What about something low tech like: the field has a grid with infrared emitters pointing up spaced let's say 6 feet apart, when changing positions extras have to keep the torch vertically for a second for it to receive its id from the local emitter, then you simply broadcast every instruction to everybody and each light knows to look only for instructions for its id
As long as you are ok with setting up the emitters on the ground and the assumption that you can "re-sync" the position at every position change, seems like it'd be fairly straightforward
edit: after a few rewatches, there's definitely a few places where (non-Chinese) human timing seems unlikely but I'm still hoping it was manual.
If each torch is designed for it's own show, then the only thing that has to be synced is the exact time to start. And in comparison, that problem is trivial to solve with known constraints (as opposed to triangulating hundreds of devices with < 1ft precision in addition to actually synchronizing the devices)