robotics-toolbox-python icon indicating copy to clipboard operation
robotics-toolbox-python copied to clipboard
verified publisher icon petercorke

Is there a way I can incorporate FK into a neural network and do back-propagation

Open Allen-Zhou729 opened this issue 2 years ago • 0 comments

Hi! Thanks for you excellent work!

Currently, I have a network which learns the IK mapping for a specified manipulator, and it outputs the joint angles for all the joints. But the performance is not good. So, I want to use the FK function (ur5.fkine()) to map the predicted joint angles to position and euler angles, and compute the L1 loss with the ground truth like the illustration below.

WechatIMG819

So, the FK process need to be able to support the loss.backward() in PyTorch. Is there a way I can make this happen?

Allen-Zhou729 avatar Jul 10 '23 12:07 Allen-Zhou729