Cosmetic change: It's ugly to have just the disembodied gripper right now. The control dynamics are the same/nearly the same as other EEF environments (regardless if they use IK solver or directly control the position of the head) but may make people believe they are different + is aesthetically worse.
I believe we cannot add the arm back to the EEF (while the arm base is anchored to the table) while keeping the EEF a freejoint as it is right now. Instead, to add back the arm, we can either:
- Add back the arm as usual and have the EEF not be a freejoint, but use a mocap + equality constraint; we control the mocap instead. Current problem is that we cannot specify an equality between the mocap and the EEF itself, because the EEF gets fused to the arm and is not considered a distinct body. The devs have this on the todo list, but have not implemented it yet: Mocap-equality issues in brax/MJX google/brax#536
- Add an IK solver and then use non-EEF control (but we need joint angles from the IK solver, not torques, or if it outputs torques, we can modify the actuators to be torque-controlled instead of being PD controllers for the target joint angle)
Cosmetic change: It's ugly to have just the disembodied gripper right now. The control dynamics are the same/nearly the same as other EEF environments (regardless if they use IK solver or directly control the position of the head) but may make people believe they are different + is aesthetically worse.
I believe we cannot add the arm back to the EEF (while the arm base is anchored to the table) while keeping the EEF a freejoint as it is right now. Instead, to add back the arm, we can either: