DOF Torque values

Hello, while attempting to evaluate the sim2real process between the panda arms in Isaac Gym and an actual franka, we noticed that there were large discrepancies in the joint torque values.

Upon closer inspection, it appears that the arms do not exhibit a significant amount of torque while at rest. This can be replicated using the franka_attractors.py example and doing the following:

  • Acquiring the dof_force tensors and refreshing them with refresh_dof_force_tensor

Observing the output of the forces tensor. During the circular motion, the forces tensors show variations which is expected, but drops to close to zero when at certain positions (presumably the initial resting position).

  • Removing the update_franka(t) step to fix the arm in position

Observing the output of the forces, they are now close to zero. This should not be the case in an actual robot due to the gravitational forces.
Is there a way to obtain the natural torque values that an arm should be experiencing at each of its joints? We have enabled gravity for the arm and etc, but so far have not found any success.

1 Like

Check this out!

They said “If you set the drive stiffness and damping to zero you should be able to do torque control. This way the position/velocity target do night fight the specified torque value.”