Do i need to add 1.02 meters to have good localization?

Technical
,
1

Me and my vision teams are debating on wheter we need to add the length of the cube grid to the april tag localization, we want to test the accuracy that limelight3 provide us.
do we need to add something? we cant point out how to check good accuracy

2

Depending on what form of localization (botpose_wpiblue, botpose_wpired, etc.) you are reading from the NetworkTables, you generally don’t have to do additional math if you match up the (0, 0) point up to what your robot’s odometry is. In my experience, I used the botpose_wpiblue and got accurate results within 2-3 inches at half-field using the Limelight 3. Hope this helps!

-Is-Han-Solo

3

thanks is han solo! we also use botpose wpi blue yet we still have around 5 inches of inaccuracy maybe because our AT arent the same as the field?

4

At what distance do you see this 5 inches of error?

5

at close distance, at long we see even more

6

One rough way to check calibration accuracy is to have 2 tags in the camera frame and do some transform math to find out the offset from one tag to the other. since you know the real number you can compare

camera returns camera_to_tag1 and camera_to_tag2 so you just do camera_to_tag1 minus camera_to_tag2

1 Like
7

You can get another view of the numbers from AdvantageScope 3D Field. Zoom in and look closely to it and you can discern less than 5 inches. I ran a couple of very different cameras into the roboRIO using WPILib math for the pose calculation and used our LL2+ (wpilib blue). Those three “robots” in AdvantageScope were essentially identical poses and I’d say they matched reality within an inch or so that I could easily measure to my AT 1.

You might try calibrating your LL if it doesn’t line up right in AdvantageScope. Other obvious errors that you likely didn’t make are incorrectly defining the position of the camera relative to the robot in LL.

What do you mean by this? The AT have to be exactly the right size and planar - no bumps, wrinkles, or curves in them. The 3D location and 3D rotation of the tags don’t matter - that’s part of the pose differential.

1 Like
8

This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.