Investigate how the local path updates after to wind shifts

[jamenkaye]

Currently, when we have a path with a corner designed for upwind sailing (first picture), and the wind direction shifts to a new direction that allows us to sail straight towards the goal (second picture), the local path doesn't update until a few minutes later when we receive a new Global path. Ideally, the local path should update automatically and more quickly. Keep in mind though, we still don't want the local path to make very small updates too frequently.

Relevant code: https://github.com/UBCSailbot/local-pathfinding/blob/099a29bb4c30e6c3ed6249f9d937d9e70230fac2/python/main_loop.py#L202

[jamenkaye]

Local pathfinding tries to update the local path, but then will come up with an identical path multiple times for at least minutes in a row. I've only seen this happen with paths that have upwind or downwind sailing, but from what I've noticed the reason for trying to recalculate the path is always costTooHigh. Sometimes when I change the wind direction, the path gets updated quickly, and sometimes it takes at least several minutes, but I haven't left any tests running longer than that.

This run was executed at 7:14 on May 30, 2022 on my computer, while I was checked out to the latest commit made to the master branch.

The attached files are a sample of the terminal output demonstrating the potential issue, as well as a screenshot of the visualiser showing the upwind sailing section that should call for a new local path.

Potential Issue.txt

[patrick-5546]

Great observations, I have a couple of comments:

• Can you detail the commands you ran so that I can reproduce it on my computer?
• Why does the path eventually update? Is it because it finally manages to find a path with a lower path cost and that doesn't require a large turn, or is it some other reason?
• What do you think the desired behaviour should be? Is a couple minutes to update the path too slow?

[jamenkaye]

• I ran roslaunch local_pathfinding all.launch, then in a terminal window I ran rosparam set /global_wind_direction_degrees 50. Use whatever angle you need to create one section of the local path to be sailing straight into the wind, then wait and observe how the local path gets updated in the visualiser.
• I think if the path contains upwind or downwind sailing at inappropriate angles, the local path should be updated in under 5-10 seconds or so. I'd be happy to ramble on about my thoughts on this, but the important part is that we are trying to balance "as fast as possible" with just enough patience to know our wind reading is accurate and we are in fact sailing upwind.

As to when it updates, here are some of my findings from testing just now:

• when the current leg of the local path directly upwind, it will detect upwind sailing. It seems like it will say it detects more than 2 seconds of upwind sailing which is grounds for an update, but then the next loop, it will only detect 1.5 seconds and not do anything. It takes a while of this before it eventually updates the local path, but when it tries to update, it seems to work correctly.
• When a future leg of the local path is upwind sailing, it tries to update because of the pathCost reason, but the path shown in the visualiser doesn't change. This can go on for several minutes, but just now I didn't test exactly how long.
• I'm not sure anymore if its just the large turn threshold getting in the way, because if the whole local path is a straight line directly into the wind, the new path should be different by more than 10 degrees. I think I should try printing out if the large turn threshold is getting activated though.

Note that just now I had to try this in my lowWindConditions branch because of some issue I had with the main branch, but previously I observed this same behaviour in the main branch as well. I could see that it wasn't in low wind conditions, so it should be operating like normal.

[jamenkaye]

Future testing/todo:

• Try to get the main branch working again so I can do testing there.
• Investigate whats the worst-case-scenario, ie the longest possible time we can be sailing in irons on the current leg of the local path without updating.
• See if we can make it update the path properly if any leg is in irons, not just the current leg.
• See if the small angle threshold is getting in the way, and modify this if needed. We don't want to be updating the local path with minor changes all the time though.