[Kevin Sevcik]

If the belt is ratcheting under load, then software solutions will only help you until your first rock wall, moat, or rampart crossing. Then you'll get the shock loading from physics instead of your PID, and you can't reprogram physics.

We've had this problem on drivetrain belts using the Vex plastic bearing blocks when the students put the cam tensioner on the non-belt side of the block. Also on other systems where structure between the pulleys is questionable. What happens is the belt loads up and the tension pulls the two pulleys closer together. This both puts an upper limit on your effective tension, and it adds misalignment into the mix which makes ratcheting more likely. I've usually diagnosed this by simply squeezing the two sides of the belt together and observing the shafts and bits of structure. If you see something moving when you squeeze the belt together, that's your weak point that you have to lock down. Your structure looks pretty solid, but it's had to tell from a video of the thing whipping around like that. If you try this and decide you need to beef up your structure, I think your best option would be something like a 1x1 aluminum tube with vex end-mount bearing gussets directly between the two shafts.

The other option is that you're simply overloading the belt. They can only transmit so much tension before ratcheting or breaking after all. I'm going to disagree with everyone else here and declare that the small pulley is NOT your problem. Or at least not your only problem. You almost can't get more wrap around that pulley, after all. The problem is that both of your pulleys are too small. Bigger pulleys mean bigger lever arms, mean less tension in the belt to transmit the required torque. I'd go with gptelli's suggestion of switching to 36T->60T or something similar. You definitely want the 60 tooth pulley on the arm, since that's what really drives the loading of the system.