Dear authors @ LeCAR-Lab stuffs,
What an amazing work！I’m integrating DIAL-MPC for robot arm pick‐and‐place tasks with both the xArm7 and xArm6. I’ve noticed a potential issue with mixed actuator types:
On quadruped or humanoid platforms like Go2 and H1, all actuators share the same motor type, so the control sequence $u$ naturally has shape $(self.nu, H_\text{sample}+1)$.
On the xArm7, however, the seven link joints use “general” actuators but the gripper is driven via a tendon actuator (we can treat it as two tendon actuators).
Because DIAL-MPC’s reverse scan and rollout assume a uniform actuator model, this mismatch degrades performance. In our tests, an xArm7 without the gripper plans a reasonable trajectory to reach the cube, but when the tendon-driven gripper is included, planning quality drops substantially, reverse-scan time balloons, and runtime delays increase.
Could we discuss how best to handle mixed actuator types in the control sequence? I’d appreciate any guidance on aligning tendon-driven actuators with DIAL-MPC’s planning framework.
Sincerely,
Chenbin Yu
ERL, UCSD
Dear authors @ LeCAR-Lab stuffs,
What an amazing work！I’m integrating DIAL-MPC for robot arm pick‐and‐place tasks with both the xArm7 and xArm6. I’ve noticed a potential issue with mixed actuator types:
On quadruped or humanoid platforms like Go2 and H1, all actuators share the same motor type, so the control sequence$u$ naturally has shape $(self.nu, H_\text{sample}+1)$ .
On the xArm7, however, the seven link joints use “general” actuators but the gripper is driven via a tendon actuator (we can treat it as two tendon actuators).
Because DIAL-MPC’s reverse scan and rollout assume a uniform actuator model, this mismatch degrades performance. In our tests, an xArm7 without the gripper plans a reasonable trajectory to reach the cube, but when the tendon-driven gripper is included, planning quality drops substantially, reverse-scan time balloons, and runtime delays increase.
Could we discuss how best to handle mixed actuator types in the control sequence? I’d appreciate any guidance on aligning tendon-driven actuators with DIAL-MPC’s planning framework.
Sincerely,
Chenbin Yu
ERL, UCSD