Hi! I’m trying to use the Jetson Nano for pose estimation on a robot using the 16h5 april-tag family. Does Issac have functionality for this april tag family? I don’t think it does, but most information I can find is from 2018 and hopefully it has been updated since then.
If it doesn’t support the 16h5 family, does anyone know of any good sources to gpu accelerate finding april tags?
Hi, Isaac SDK supports 36h11 tags, and for Jetson Nano there is an example called “follow_me” for the Kaya robot that uses these tags with an Intel RealSense D435 camera. We have used it successfully. We also have another Kaya that has a toy Lidar and standard wide angle camera (not Intel RealSense) but have not tried the follow_me on that Kaya robot yet.
The description for 16h5 says it uses fewer bits and might have higher error rate, so perhaps 36h11 would be more suitable for your project.
See here:
https://docs.nvidia.com/isaac/doc/tutorials/kaya_software.html#the-follow-me-application
SIX years later, I petitioned for a more serious consideration of this in this thread:
It’s kinda silly that we can’t do GPU accelerated 16h5 tag handling in 2023.
We have not looked much at adding support for more tag families but we’ll try to get this prioritized.
I want to bump this as well, I’m interested in using other tags in my work at Carnegie Mellon
@avrickaltmann Did you ever find a solution for this? FRC season is upon us again, and I’ve just ordered some Orin Nanos for our vision, including (hopefully) 16H5 detection.
@Phrogz Fortunately, FRC is using 36h11 tags this year which, if I recall correctly, Issac has support for. I did however move from Issac to openCV last year though because I did not end up needing the GPU performance boost.
For FRC specifically, the photonvision library is a great resource as well. https://photonvision.org/