Joint simulation is poor

[nickswalker]

Here's a video of how the current simulation behaves when trying to pick up a coke can:
https://drive.google.com/file/d/145Gu7RzlPUGtJEBlLAHFDucbPxuanf0W/view?usp=sharing
(A solid antipodal grasp causes large forces in the gripper and the object. The object pops out of the gripper and the robot shudders).

This behavior is due to the incomplete specification of the simulated motor controller parameters. Without gains, no actual simulation of the joints is occurring. Instead, the default gazebo hardware plugin is ignoring dynamics and forcing the joints into the target positions.

I have a near-complete patch that resolves this behavior, but I haven't had luck stabilizing the wheel joints. I'm happy to start a PR with these changes (and maybe someone with more Gazebo expertise can figure out what's going on with the wheels!), but the plausible simulation may disrupt use cases that depend on the current behavior, so I thought it would be good to encourage discussion from any other simulation users first.

[hello-binit]

Thank you @nickswalker for filing this issue and starting on a patch! @vatanaksoytezer is the author of much of the Stretch Gazebo, so he would have the best idea on what could be causing issues with the simulated wheels.

[vatanaksoytezer]

Hello @nickswalker, thanks a lot for the issue!

I have a near-complete patch that resolves this behavior, but I haven't had luck stabilizing the wheel joints. I'm happy to start a PR with these changes (and maybe someone with more Gazebo expertise can figure out what's going on with the wheels!), but the plausible simulation may disrupt use cases that depend on the current behavior, so I thought it would be good to encourage discussion from any other simulation users first.

If you can send a PR, we can move the discussions to it and stabilize the grasping behavior as quickly as possible, as this should be a core feature we would like to have.