Req: Help getting started with pose estimation

Robotics & Edge Computing Jetson & Embedded Systems Jetson Orin Nano
1

Hi. I purchased a Jetson Orin Nano for my son who participates in the First Robotics Competition.

He wishes to learn to do pose estimation of objects from CAD drawings.

Although we have taken some of the Fast.AI class and have managed to use the FAST.AI image recognition library, we’re still pretty new to all of this.

Can someone recommend a path forward? We tried to look at “Pose Estimation using CNNs with ISAAC 2.0”, but it seems to require the Carter robot to do the tutorials, which we don’t have.

I see there is also a YOLOV8, but a poster said it doesn’t run well on Orin Nano. (Perhaps that’s a resolved issue?) Also, we’re not sure if YOLOv8 can be used to detect only what we have trained and not the rest of the things in its library.

Is YOLOV8 a good place to try to get to, or would something else be better? And where would be a good tutorial be to start learning to use our Jetson Nano?

Thanks.

3

Hi,

Do you have an example of CAD drawings?
We do have some post estimation but it is based on real-world image.

Do you mean human pose?
Thanks.

4

Every year is different, but typically the things we would want to recognize are cubes, cones, balls, and target structures, such as hoops, towers, etc.

This page has a lot of CAD drawings that include the kinds of things we would want to know the pose of:

Playing Field | FIRST (firstinspires.org)

5

Hi,

Sorry that we don’t have an example for the object pose estimation.

You can try to find some model that are based on TensorFlow or PyTorch.
These two frameworks are supported on the Jetson.
To convert these models to TensorRT is also supported.

Thanks.

7

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.