AprilTag example is broken on Rio

[PeterJohnson]

The AprilTag vision example is broken on Rio as it allocates far too much memory for 36h11 by using hamming distance 3.

Hamming 1 allocates (587 + 587 * 36) * 3 * 12 = 781KB
Hamming 2 allocates (587 + 587 * 36 + 587 * 36 * 35) * 3 * 12 = 27.4 MB
Hamming 3 allocates (587 + 587 * 36 + 587 * 36 * 35 + 587 * 36 * 35 * 34) * 3 * 12 = 932 MB

(reference: apriltag.c source code in quick_decode_init)

The Rio 1 only has 256 MB of total memory. The Rio 2 only has 512 MB.

It should be changed to use hamming 2 on Rio 2 and hamming 1 on Rio 1.

[HarryXChen3]

Is this the relevant line?

allwpilib/wpilibjExamples/src/main/java/edu/wpi/first/wpilibj/examples/apriltagsvision/Robot.java

Line 43 in 74f6486

detector.addFamily("tag36h11", 3);

I assume the fix would just be to use RobotBase.getRuntimeType() and change the hamming distance accordingly? What should be hamming distance be in simulation? (does simulation even matter for this?)

[PeterJohnson]

Yes, that’s the relevant line. It’s probably better to not have different behavior on Rio 1 vs Rio 2, so it should probably just be changed to 1, with a comment that 2 is feasible on Rio 2.