Produce more accurate odometer fixes by averaging multiple GNSS poses while standing

[rodja]

As an alternative to a Kalman Filter, this PR leverages the fact that we are standing still while the implement is working:
We can collect multiple GNSS measurements and average them for a odometer pose update just before we start driving again. The new code also ensures that the navigation pauses it's advancements shortly to gather a some gps measurements every 10 seconds for scenarios where the implement does not request stopping on their own. And the AutomationWatcher was updated to generate GNSS pose updates if no automation is running and robot is standing.

ToDos

• merge Remove individual field references in favour of a simplified "global" reference for all coordinates #129 into main and then main into this branch; otherwise some coordinate changes will not have the expected effect
• what do we do if we have no gnss signal?
• test on robot
• check if Driving timeout #128 is still necessary or if we can close Driving Timeout #126
• check if Tornado x-offset #127 can be closed
• why is assert not self.is_paused raised in update_robot_pose -> missing super calls

[pascalzauberzeug]

Code looks good, looking forward to testing it

[rodja]

Today, when testing with a real robot I discovered that we run into an endless loop if we do not have a RTK fix. This can happen indoors, when no internet connection is available, when SAPOS is not correctly configured or a thousand other causes. How should we handle this?

[pascalzauberzeug]

@rodja do you mean the while-loop in update_robot_pose?

[rodja]

Yes

[pascalzauberzeug]

The await self.gnss.update_robot_pose() call should have a timeout and either we just continue to drive without an updated pose and warn the user about it or stop the automation.
I think the first option is better for our current use cases

[rodja]

I agree. But before this PR, the default behaviour was simply to keep driving with odometry. Let's discuss this tomorrow in our all-hands meeting.

[rodja]

As discussed we keep the old behaviour for now which does not require GNSS to be active. This allows us to drive by visual cues only.
By setting system.gnss.ensure_gnss = True this behaviour can later be changed by navigation strategies. The slippage-tests need it already. @pascalzauberzeug I think the code is ready for another real world test.

[pascalzauberzeug]

The worked today on U4/RB28, but one time there was the following error where the automation was stopped and couldn't be started again without restarting the software.

rosys_1       | 2024-07-29 18:39:00.551 [INFO] rosys/rosys.py:69: automation stopped because an exception occurred in an automation
rosys_1       | 2024-07-29 18:39:00.552 [INFO] rosys/rosys.py:69: automation failed
rosys_1       | 2024-07-29 18:39:00.575 [ERROR] nicegui/app/app.py:119:
rosys_1       | Traceback (most recent call last):
rosys_1       |   File "/app/nicegui/background_tasks.py", line 52, in _handle_task_result
rosys_1       |     task.result()
rosys_1       |   File "/app/rosys/automation/automation.py", line 56, in __await__
rosys_1       |     signal = send(message)
rosys_1       |              ^^^^^^^^^^^^^
rosys_1       |   File "/app/field_friend/automations/navigation/navigation.py", line 50, in start
rosys_1       |     await self.gnss.update_robot_pose()
rosys_1       |   File "/app/field_friend/localization/gnss.py", line 119, in update_robot_pose
rosys_1       |     assert not self.is_paused
rosys_1       | AssertionError

[rodja]

The tests with U4 worked very well after we fixed the work_x parameter. There also was an issue when pressing the stop button because of missing super() calls. This is now fixed. @pascalzauberzeug could you review the last changes? I think it's ready to merge!

[pascalzauberzeug]

@rodja the field navigation tests fail after 24d0da5.
this assert failsassert system.field_navigation.automation_watcher.field_watch_active in test_approaching_first_row and assert not system.automator.is_running in test_approaching_first_row_when_outside_of_field.

I narrowed it down to the await self.implement.activate() in FollowCropsNavigation.prepare()

[pascalzauberzeug]

@rodja

field_friend/field_friend/automations/implements/recorder.py

Line 23 in 58a7354

await rosys.sleep(3) # NOTE: we wait for the camera to adjust

The test doesn't wait for this sleep. We could forward for x seconds or move await super().activate() to the end of the function and then forward until the implement is active
await forward(until=lambda: system.current_implement.is_active)

[pascalzauberzeug]

Works again, will merge now!