Question about the estimated heading in GPS mode

[gponcon]

This wont affect gps heading at all.   It is only the compass offset since pypilot doesn't know which way the imu is facing.

Indeed.

That is a really bad sign.   I threw away more than 100 of the mpu9250/mpu925x because so many failed my basic tests.  Since using the much more expensive icm20948 (which has no better sensors assuming they work)  I do not have so many bad chips.   It really depends on where you get the sensors.

This is interesting. As soon as I can I will try the ICM-20948. At the time I based it on this information: https://pypilot.org/wiki/doku.php?id=inertial_sensors.

It basically means
1)  the accelerometer bias is high.  You can calibrate this out to some degree, but if its too far out of range, the +- 2g will saturate, and the software would have to use +-4 g instead which is not implemented and would degrade performance somewhat.   I simply dont use sensors more then 0.3g bias.

2)   the gyro bias is very high.   Some sensors have really high gyro bias.

Also...   some sensors have a failed axis.  So only read zero for one or more of the sensors.   pypilot doesn't do a great job of detecting every sort of possible screwed up sensor.   It is on my todo list, to make it work if the compass sensors failed, or make it warn you if one axis failed or what.. but its not implemented.

You should really verify your intertial sensors.   You can use the pypilot_scope to plot the accelerometer, and gyros, and (check gyro bias too) to ensure they are sane.   If you can check the boat plot by running pypilot_calibration it should not randomly spin at startup as you suggest it might.

Finally,  check the calibration plots.   These are available from the web, opencpn plugin, or pypilot_calibration now.   Please verify the accelerometer and compass properly calibrate.

I'm currently at anchor with a slight swell and after a calm sail during which everything went very well. I was also able to use the compass mode to optimise my course through a cross-current. The values found with pypilot_scope and pypilot_calibration seem consistent.

I am not yet very comfortable interpreting calibration plots but the messages "need more range" and "0d fit, insufficient data" are very common... I guess this is not a good sign.

A sensor failure detection system would be interesting, especially to check that the GPS heading variation matches the accelerometer variation. But it is also the responsibility of the users to use qualitative components .

The problem is...  pypilot's gps mode really is a compass mode. It is steering to compass course, and very slowly updating this internal compass course to steer to make the gps course desired.    gps can be very noisy and especially at lower speeds it would give very bad performance to not use the compass.

This explains the problem, which is seen when the sensor heading is constantly "turning". As the compass heading is constantly turning right, it is logical that pypilot wants to restore the course by turning left, even if the GPS heading is getting worse and worse...

I have a work-in-progress but not enabled "gps pilot"  which does more what you might have expected and purposefully does not use the compass instead relying on the gps and gyros.    The idea was for this to work around magnetic distortions like sunken boats or metal bridges, but it would also enable better use with a faulty compass.   It is just not completed let alone tested, so its not possible to use, and I would expect poor performance at low speeds.

Even though poor performance is to be expected at low speeds, this option would be useful to compensate for possible accelerometer failures. There is also the added benefit of being able to travel with the pilot in high latitudes where the compass no longer works.

(...) I can only assume that the IMU is not really working properly, so giving extremely erroneous errors that break the gps filter. (...)If the course it reports is all over the place and wrong, that would explain the issue. You will need to get pypilot working in compass mode to really fix this.

Today I'm going to be sailing in stronger conditions, testing and trying to see what happens if the pilot changes the course. I'll keep you informed. Thank you very much for all these precisions.