I was also a bit surprised that more primitive CV wasn't used to track the ball--he had to freeze the ball and then machine holes into the ball to insert tracking objects. I would have thought that image recognition could have solved the problem without the trackers. Maybe the issue was that this requires very low latency?
The 'cable slack takeup' devices are also unnecessary because that can be done in software - you can put a virtual spring in the motor control feedback loop to make it act as springy or rigid as you like.
I can imagine the need for both came about from previous (unseen) design iterations that had low frame rate visual feedback and motors without feedback at all.
Tape measures are bendy and would require keeping / knowing the right angles even if the walls and floor are not completely straight. Then you get measurement error on top (both in the wall plane coordinates and how far does the attachment bit sticks out) that skews everything a bit.