Limelight 3 AprilTag Pose Estimation and/or Calibration Issues

We’ve been trying to use the pose estimation from the 2024 AprilTags with our Limelight 3, but we’ve been seeing a discrepancy from the botpose estimation provided by Limelight and our direct measurement to the edge of the field.

Depending on our distance to the AprilTag, the pose estimation error also seems to grow in a non-linear pattern, with a very large error towards the edge of the limelight’s view of the tags.

The increase in error depending on our distance to the tag led us to believe that the problem lies in our camera’s calibration. So we explored using Limelight’s built-in ChAruCo Camera Calibration, but in our numerous repeated calibrations, our reprojection error has been consistently above 50, with all of our results ending up being even more unreliable than the default calibration.

We’ve checked our other pipeline settings numerous times and checked our AprilTag sizes and positions to ensure there are no physical errors, but we haven’t been able to come up with a solution yet. Any insight into this issue would be greatly appreciated.

I’ve also linked 3 posts that have very similar issues to ours, but I haven’t seen any good solutions provided so far.
https://www.chiefdelphi.com/t/limelight-field-pose-estimation-discrepancy/452331
https://www.chiefdelphi.com/t/limelight-3-calibration-issues/450714
https://www.chiefdelphi.com/t/limelight-3-charuco-camera-calibration-issues/452986

We calibrated multiple times as well, and eventually got it right. It took more than 50 snapshots for one calibration to get decent results.
We use LL2+, but I think the idea is the same.
Also make sure your test pattern is printed with the right size.

1 Like

Ok, we’ll try calibrating more. We also checked our test board with calipers before using it so we’re pretty sure that isn’t the issue. What was your final reprojection error or pose estimation error with the better results?

Same issues here. Fixed it today by switching to photonvision. Read more about it if you want on [
our build thread:

1 Like

The pose estimation error for us is at most 0.2m at the 3.5m distance (we will also have it looking sort-of UP so we can see Apriltags even if we’re against the Amp). It’s acceptable for our purposes. LL did not work that well for the long distances (and we did not need it to as we plan to use LL as a short-range camera and have a separate long-range one with PhotonVision. There we have about the same max estimation error at 7m.