Update Limelight Positions on compbot

[noahlinton]

We have added a second limelight to the back of the robot. The cameras are angled out by 10 degrees.

We are still trying to tune their position, so the data is not very accurate. We may need to increase the yaw angle even more

@WyattDube @SteveJacka @Ben-Alderson What should we do next to improve this performance?

[noahlinton]

Updated tasks:

• [x] Add specific shuffleboard layout for camera tuning
• [ ] Perform lens calibration on both limelight camera's - ensure to follow best practices
• [x] setup infront of target with laser level
• [x] Measure pitch of camera using phone level
• [x] check field layout in photonvision
• [x] Check pose estimate on shuffle board coming from cameras and adjust per spreadsheet.

[noahlinton]

@Ben-Alderson did we apply the instructions here? Looking at 2023 code, it looks like this is different than last year. Specifically, looks like we need to give the limelight the field layout. This is based on our code calling MULTI_TAG_PNP_ON_COPROCESSOR in Vision.java

https://docs.photonvision.org/en/latest/docs/apriltag-pipelines/multitag.html#multitag-localization

[Ben-Alderson]

Yes looks like you need to give it a field layout. Perhaps that's why the custom JSON for the practice field didn't work very well. It's another thing we need to remember to reset for events.

[noahlinton]

Hmm okay. We will be playing with this on Wednesday at the field. Feels like a black box a little. Hoping we can sort it out.

[noahlinton]

[Ben-Alderson]

The logic for scaling by distance is already integrated in Vision::getEstimationStdDevs.

The base standard deviation that function uses can be tweaked here in Constants.java. Larger numbers mean we "trust" the values less. So a bigger number will mean it takes longer for the robot to get to its true position, but it won't jump around as much.

[noahlinton]

@Ben-Alderson any advice on how to actually tune these values? I am reading through the classes to understand how they are implemented, and I see we have two cases, single target, and multi target. And I see that if the distance is greater than 4, we use "max values" (Min trust).

It looks like we could also maybe create a stdDev for use in auto?

Do we just guess and check these numbers? Do we look for the pose calming down?

[Ben-Alderson]

@noahlinton Yeah there's not really an objective method for optimizing these values. I would just mess around with it until it doesn't jump around much but still gets to the real position quickly enough.

Not sure why the stdDev would be different in auto?

[noahlinton]

Okay will do! Dont have a reason for being different in auto. Just thinking