I really think we should use the 36h11 apriltag family. While testing vision I’ve noticed that the more detailed and complicated 36h11 tag got better results. Does anyone know why they changed to 16h5. Tell me if you agree
3 Likes
A bunch of discussion in the thread from the announcement: [FRC Blog] 2023 Approved Devices, Rules Preview, and Vision Target Update
The main advantage of 16h5 seems to be longer detection distance with the same image resolution, resulting in better frame rates.
I think it’s a worthwhile question to revisit though as we’ve now had build season and quite a few events for teams to test the available solutions (e.g. PhotonVision, LimeLight, home-grown) for 16h5 on real fields. Was 16h5 good enough or would 36h11 have made a difference (either good or bad)?
4 Likes
Personally we weren’t really using the tags as a long rage tool and more as a calibration/fiducial system and as a way to line up. The 36h11 we’re less gittery with less calibration and fine tuning
As an additional data point, I’m not sure the farther detection distance gets us anything. After spending some time looking at the vision estimates we gather from looking across the field to the other alliance’s apriltags, every pose estimate gets the distance correct, but the position relative to the tag is just wonky. It’s either high in the air, far off to one side or the other, or under the field.
We’re using a LL3 on the latest image, 960x720@22fps. Detection config is otherwise default, and we utilize megabotpose rather than the estimate from an individual tag. We reject any vision estimates that include tags >7m away from the camera.
I’ll probably provide that dataset sometime this week to demonstrate what we’ve observed.
2 Likes
With 36h11, you have the option to use the really affordable and pretty capable HuskyLens. I could see this being a great thing, potentially for FTC as well.
It’s not a Limelight, but it offers pretty decent tag tracking at a pretty low barrier to entry.
4 Likes
36h11 is much more stable, but the 25% detection distance loss might be a pain. That said, as it stands, I would take it if we got lower ambiguity or higher accuracy. I haven’t tested those values, however.
640x480 can detect a tag from 20-30ft away with 16h5. That’s a big upgrade from 15-16ft with 36h11, because it matters in many launching games.
2 Likes
I’d like to see more tags being used on the field first.
6 Likes
Note this can be solved with more fiducials on the field too which 36h11 is better at since its a larger family.
2 Likes
36h11.
More of them.
Marnack has spoken.
23 Likes
A big problem we faced was poor results due to robots blocking what few AprilTags we could see, resulting in us relying on single-tag pose estimation (which is pretty awful compared to SolvePNP). This wasn’t a problem in the shop but resulted in our robot flying around in auton during practice day, so we ultimately had to remove it.
This is the answer. We need more tags.
Indeed, you might not have, but multiple other teams were attempting to do so. Not every game will necessarily want close-range targeting.
I found it quite surprising the few amount of April tags on the field this year. I expected at least a a few on the plastic side walls to help with better localization.
1 Like
I feel like this year’s apriltags were intended somewhat like “check out these new targets–now with fewer blinding green lights!” and localization was not considered sufficiently as a potential use case and therefore got neglected. I would guess that’s why there’s apriltags near scoring and pickup locations but not anywhere else.
3 Likes
Extremely true, I overlooked this. I’ve changed my mind, 36h11 with extra tags would be much better, especially for “megatag” detection.
1 Like
16 Likes
more
more
9 Likes
So I think the #1 priority here should be advocating for more tags. Having 2 tags in view makes the pose significantly more accurate and reliable.
While 16h5 still has some room to expand ideally FIRST would use even more than it could support, this is made possible by 36h11.
However, if FIRST insists on still using as few tags as possible I guess it would be fine to stay with 16h5.
1 Like
First has made it clear that they don’t want to increase the amount of tags, or use anywhere more than 30 (planned significantly less). I plan on aggregating apriltag feedback to send to FIRST and to improve documentation.
2 Likes
FIRST would do well to remember that they serve teams and not vendors with this.
14 Likes
“gerthworm’s apriltag and apriltag accessories online emporium” stands to profit greatly from a many-tag setup.
4 Likes