I have been working with the 0.56 release this summer and have found the following anomaly. This system will hold a good course for a period of time which may be minutes or hours. Eventually, the course begins to drift to port. The commanded course remains the same and the system is actively steering but it progressively goes further and further off course. I get these results whether in nav, gps, or compass mode. When I moved to 0.56, I carefully copied all the gain settings from 0.24 which continues to work well.
I like the new features in 0.56 but holding a course is more important. Has anyone else experienced this or have any insights or suggestions?
Are you running on the same hardware? Did you revert back to 0.24 and the problem stopped??
How does the commanded course compare to the reported course?
Did you try locking the compass calibration once completing a circle. maybe it performed an update that was actually wrong.
(2025-07-05, 09:18 PM)seandepagnier Wrote: [ -> ]Are you running on the same hardware? Did you revert back to 0.24 and the problem stopped??
How does the commanded course compare to the reported course?
Did you try locking the compass calibration once completing a circle. maybe it performed an update that was actually wrong.
Yes, same hardware. The old version still works. The compass calibration has never looked right with either version. It always says it has too few points.
this is most likely the issue.
You need to ensure the compass calibration is unlocked and complete a full 360. At this point the calibration should update with new bias and the points fall on the sphere. Now the heading displayed should change by 90 degrees or so when you turn 90 degrees and make sense (it may still have an offset you can manually enter)
Once this looks good I would try locking and trials. This is true for both versions.
Before calibrating be sure to level the sensors to the boat.
(2025-07-06, 12:00 AM)seandepagnier Wrote: [ -> ]this is most likely the issue.
You need to ensure the compass calibration is unlocked and complete a full 360. At this point the calibration should update with new bias and the points fall on the sphere. Now the heading displayed should change by 90 degrees or so when you turn 90 degrees and make sense (it may still have an offset you can manually enter)
Once this looks good I would try locking and trials. This is true for both versions.
Before calibrating be sure to level the sensors to the boat.
I have never been able to get a full compass calibration, even with the unit dismounted and turned by hand. Could this be a hardware problem? In spite of this, 0.24 works well so I have not been too worried.
what does it show for compass bias figures? Maybe a hardware problem but can you even maybe plot the data using the pypilot scope?
It can be difficult to calibrate dismounted. A likely reason one version may work but not the next would be alignment. Be sure to level the sensors once mounted to the boat and the boat is sitting still.
Another issue is possible magnetic interference from metal objects. Is it possible somehow this could have caused issues testing version 0.56 and it didnt occur when testing 0.24 making a false conclusion?
I am a little confused how the older version would work but not the newer, the logic is not much changed in steering only interface improvements, but if you can verify this it would be helpful. It seems like your observations could be correct but are not scientific. Myself and several other people have verified the new version as working and there are actually still a few minor bugs (some fixed in github but not 0.56) but nothing related to steering as you describe.
(2025-07-08, 04:22 AM)seandepagnier Wrote: [ -> ]what does it show for compass bias figures? Maybe a hardware problem but can you even maybe plot the data using the pypilot scope?
It can be difficult to calibrate dismounted. A likely reason one version may work but not the next would be alignment. Be sure to level the sensors once mounted to the boat and the boat is sitting still.
Another issue is possible magnetic interference from metal objects. Is it possible somehow this could have caused issues testing version 0.56 and it didnt occur when testing 0.24 making a false conclusion?
I am a little confused how the older version would work but not the newer, the logic is not much changed in steering only interface improvements, but if you can verify this it would be helpful. It seems like your observations could be correct but are not scientific. Myself and several other people have verified the new version as working and there are actually still a few minor bugs (some fixed in github but not 0.56) but nothing related to steering as you describe.
I have never looked at compass bias or run pypilot scope but I will. Right now I am try to fix my gearbox so am somewhat preoccupied. 0.56 can run perfectly for long periods and then start to turn. Going to standby then re-engaging often (temporarily) gets it back on track. I will try to send more data when I can get it.
If I wanted to buy a new IMU module, what part number should I look for?
(2025-07-09, 06:46 AM)oceandrean Wrote: [ -> ] (2025-07-08, 04:22 AM)seandepagnier Wrote: [ -> ]what does it show for compass bias figures? Maybe a hardware problem but can you even maybe plot the data using the pypilot scope?
It can be difficult to calibrate dismounted. A likely reason one version may work but not the next would be alignment. Be sure to level the sensors once mounted to the boat and the boat is sitting still.
Another issue is possible magnetic interference from metal objects. Is it possible somehow this could have caused issues testing version 0.56 and it didnt occur when testing 0.24 making a false conclusion?
I am a little confused how the older version would work but not the newer, the logic is not much changed in steering only interface improvements, but if you can verify this it would be helpful. It seems like your observations could be correct but are not scientific. Myself and several other people have verified the new version as working and there are actually still a few minor bugs (some fixed in github but not 0.56) but nothing related to steering as you describe.
I have never looked at compass bias or run pypilot scope but I will. Right now I am try to fix my gearbox so am somewhat preoccupied. 0.56 can run perfectly for long periods and then start to turn. Going to standby then re-engaging often (temporarily) gets it back on track. I will try to send more data when I can get it.
If I wanted to buy a new IMU module, what part number should I look for?
I have installed the pypilot scripts. What would yo like me to plot and under what conditions?
I can confirm the problems mentioned above from oceandrean. Latest version unusable and same problem here. Seams to have something to do with imu (compass distortion). Wasted time and money trying new imu and extender with no effort. So far version tinypilot_2021_11_16.img working for keeping course but not storing settings. Only possible by writing to config file.
Can someone give me the most reliable and usable version of tinypilot. Unfortunately I deleted older downloads. Just will try on sea tinypilot_beta_2024_03_25 tomorrow.
Thanks for the feedback.
So you are saying the version:
7_17 from
https://pypilot.org/images/
gives you issues to confirm?
I did sail with this version and had no problems and have also received recent feedback from users who claimed it worked well. My best guess is that the compass is re-calibrating while underway and this causes issues. I have seen this problem in waves where it thinks it has enough data for a calibration update but really should not, and it something of a bug, but the solution is not obvious.
The clearest way to work around this is by completing a circle and ensuring the compass reading is correct, then locking the calibration.
If it is something else I really would like to know.
ocean dream claims he never got a compass calibration. It may be that the older version worked because the default bias was already close enough to whatever sensors he had, where the new version it is actually zero bias. This is the only explanation I can give. I would ensure you get a calibration update after completing a circle then lock calibration.
If you see this issue again, can you please notice if the displayed course is near the commanded course, or not? I am trying to determine if you notice the issue because the calculated heading is wrong or some other bug where it is accumulating error. Also maybe set the I gain to 0 if it is not already.