`calibrate()` panics: attempting to subtract overflow

[nguterresn]

Hey there!

I'm trying to use this crate with my esp32c6 and esp-hal 1.0.0-beta1.

Whenever I try to calibrate using the calibrate method, it panics.
I have no problem using the deprecated alternatives. Is this a known issue?

[jrjdavidson]

Same problem here. I have two different types of mpu6050s - one type panics on calibration, the other is fine.

[ERROR] panicked at C:\Users\jrd92\.cargo\registry\src\index.crates.io-1949cf8c6b5b557f\mpu6050-dmp-0.6.0\src\calibration.rs:97:13:
attempt to subtract with overflow

This sensor type panics:

This sensor type calibrates without issues:

[barafael]

Thanks for reporting and detailing the issue! I'll be looking into it see what I can do.
Is it possible the mpu6050 are not original parts? Regardless the driver should not panic of course.

Do the deprecated calibration functions work for you, too?

[nguterresn]

interesting @jrjdavidson
I'll try with others I have around and report back. I bought mine from a random supplier from AliExpress, so probably a clone or similar.

[jrjdavidson]

Not sure about the origin of the parts, sorry.

Do the deprecated calibration functions work for you, too?

Not sure what you mean here- if you let me know i can test it later.

[jrjdavidson]

did a bit more poking around and added some logs to see what is going on. It appears that the offset on the accel never gets set, meaning the values don't converge. The gyro values appear to work fine.

[INFO ] Current mean: 376, current offset: -32768
[INFO ] Calibrating accel_y
[INFO ] Current mean: -559, current offset: -14629
[INFO ] Calibrating accel_z
[INFO ] Current mean: -30483, current offset: 32767
[INFO ] Calibrating accel_x
[INFO ] Current mean: 371, current offset: -32768
[INFO ] Calibrating accel_y
[INFO ] Current mean: -555, current offset: -14573
[INFO ] Calibrating accel_z
[INFO ] Current mean: -30492, current offset: 32767
[INFO ] Calibrating accel_x
[INFO ] Current mean: 376, current offset: -32768
[INFO ] Calibrating accel_y
[INFO ] Current mean: -554, current offset: -14517
[INFO ] Calibrating accel_z
[INFO ] Current mean: -30474, current offset: 32767

Note I did change the offset calculation to prevent the overflow:

            let adjustment = (current_mean / 10) + current_mean.signum();
            current_offset.saturating_sub(adjustment)

[jrjdavidson]

I can confirm, regardless of the offset, the values read by the sensor remain unchanged. Here is an extract that starts from 0 offset for accel and you can see they remain unaffected. I'm not sure what the cause could be here - hoping someone else can suggest something..

[INFO ] Calibrating Sensor
[INFO ] Current accel calibration: Accel { x: 0, y: 0, z: 0 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12428, current offset: 0
[INFO ] Calibrating accel_y
[INFO ] Current mean: 664, current offset: 0
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7372, current offset: 0
[INFO ] Calibrating gyro_y
[INFO ] Current mean: 80, current offset: 21
[INFO ] Current accel calibration: Accel { x: -1243, y: -67, z: -738 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12428, current offset: -1243
[INFO ] Calibrating accel_y
[INFO ] Current mean: 657, current offset: -67
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7378, current offset: -738
[INFO ] Calibrating gyro_y
[INFO ] Current mean: 148, current offset: 12
[INFO ] Current accel calibration: Accel { x: -2486, y: -133, z: -1476 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12430, current offset: -2486
[INFO ] Calibrating accel_y
[INFO ] Current mean: 655, current offset: -133
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7376, current offset: -1476
[INFO ] Calibrating gyro_y
[INFO ] Current mean: 18, current offset: -3
[INFO ] Current accel calibration: Accel { x: -3730, y: -199, z: -2214 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12432, current offset: -3730
[INFO ] Calibrating accel_y
[INFO ] Current mean: 655, current offset: -199
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7375, current offset: -2214
[INFO ] Calibrating gyro_y
[INFO ] Current mean: 12, current offset: -5
[INFO ] Current accel calibration: Accel { x: -4974, y: -265, z: -2952 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12432, current offset: -4974
[INFO ] Calibrating accel_y
[INFO ] Current mean: 652, current offset: -265
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7376, current offset: -2952
[INFO ] Calibrating gyro_y
[INFO ] Current mean: 5, current offset: -7
[INFO ] Current accel calibration: Accel { x: -6218, y: -331, z: -3690 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12431, current offset: -6218
[INFO ] Calibrating accel_y
[INFO ] Current mean: 655, current offset: -331
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7388, current offset: -3690
[INFO ] Current accel calibration: Accel { x: -7462, y: -397, z: -4429 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12423, current offset: -7462
[INFO ] Calibrating accel_y
[INFO ] Current mean: 655, current offset: -397
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7375, current offset: -4429
[INFO ] Current accel calibration: Accel { x: -8705, y: -463, z: -5167 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12434, current offset: -8705
[INFO ] Calibrating accel_y
[INFO ] Current mean: 658, current offset: -463
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7375, current offset: -5167
[INFO ] Current accel calibration: Accel { x: -9949, y: -529, z: -5905 }
[INFO ] Calibrating accel_x
[INFO ] Current mean: 12426, current offset: -9949
[INFO ] Calibrating accel_y
[INFO ] Current mean: 656, current offset: -529
[INFO ] Calibrating accel_z
[INFO ] Current mean: 7388, current offset: -5905
[INFO ] Current accel calibration: Accel { x: -11192, y: -595, z: -6644 }

[1-rafael-1]

the values never converging is what i experience when using a mpu-6500 as opposed to a mpu-6050. On the latter i never ran into an overflow or values not converging.... so my last fix on the overflow was thrown in as a best effort and we may be overlooking things.

[1-rafael-1]

@barafael i have my laptop but no hardware.... #22 is a 100% AI fix which I cannot test IRL right now. But since I never had the issue on actual hardware, maybe @nguterresn could give it a shot?

[barafael]

@1-rafael-1 thanks! @nguterresn would you mind testing the current main branch of this repo?

[1-rafael-1]

or maybe @jrjdavidson could test on main branch?

[jrjdavidson]

#22 is the same fix as I mentioned above using saturating_sub() - it prevents the panic but doesn't address the issue. For some reason, the offset doesn't apply to the readings on those chips. The offset keeps growing in an attempt to reduce the mean value, ending in the overflow.
E.g. if you take the logs above and only look at accel_x values:

[INFO ] Current mean: 12428, current offset: 0
[INFO ] Current mean: 12428, current offset: -1243
[INFO ] Current mean: 12430, current offset: -2486

You can see that the mean value remains roughly the same while the offset grows.

Is it possible that the offending mpu has a different register for calibration values? Is it possible that the it does not implement calibration? I think if we did prevent the panic using saturating_sub(), we should also have some kind of error mechanism that checks if the offset is successfully applied so that the program can continue running if calibration fails.

[nguterresn]

@jrjdavidson I had the same issue as you — some MPU6050s were not working. I could already tell by the return output of the temperature which was around 50C degrees...

I'd say those who trigger the panic should be considered as a faulty chips, or clones. I'd advise not to use them.

[jrjdavidson]

Hi all,
I found the issue for the sensors I have.

In the document below, the register for the accel offsets are noted as being different for the MPU6500 compared to the MPU6050.
https://www.digikey.com/Site/Global/Layouts/DownloadPdf.ashx?pdfUrl=BA826F044684489C833D66EBE99F7DB1

indeed, when checking the whoami register, the sensor that panics returns 0x70 (corresponding to MPU6500) instead of 0x68 like the other sensors, adn this despite being clearly marked as MPU6050! I ran a quick test with the MPU6500 accel registers, and indeed the x,and y accel calibrated fine, but for some reason the z axis still didn't calibrate.

Perhaps a whoami check could be included when initialising the sensor?

[barafael]

@nguterresn in case you still have the hardware, could you check the whoami register as indicated in the previous comment?