search

pybricks / support

Pybricks support and general discussion
MIT License
109 stars 7 forks source link

[Bug] Bad internal rounding causes gyro controller to be off by 0.3%. #1886

Closed laurensvalk closed 1 month ago

laurensvalk commented 1 month ago

Describe the bug / To reproduce

round(hub.imu.heading()) is supposed to be equal to robot.angle() when the gyro is used, but was not the case.

image

For example, 720.3 should round to 720, not 722.

This was linearly increasing for bigger rotations, adding to the apparent drift when there wasn't any.

Expected behavior Round properly. And then later switch to floats in #1844.

We need to fix the rounding either way because this value is still used by the internal controllers, causing things like turn(360) be off to up to a degree.

Additional context This issue has already been debugged, but opening this issue serves as a documentation of the issue and its fix. Fix incoming shortly.

laurensvalk commented 1 month ago

I think this could have subtlety affected other gyro issues reported by @afarago and others, but I can't find the exact one right now.

Basically, you could end up in a situation where you applied turn(360) but hub.imu.heading() would report 358 or 359.

Note that this is strictly separate from calibration, because calibrated or not, turn and heading should ultimately in exactly the same thing, but they were not for some drivebase configurations. (This is different from where the robot actually ends up in reality, which is affected by calibration.)

After this fix, heading and turn use the same values as intended.