Why use the g1_29dof_hard_waist configuration in the training code?

[Kennyp-Chen]

Why use the g1_29dof_hard_waist configuration? Does it offer a greater advantage than the sim2sim checkpoint version of training?

[BlackOtters]

if you use g1_29dof the sim2sim performance is not good

[Kennyp-Chen]

if you use g1_29dof the sim2sim performance is not good

Thank you for your reply. Could you elaborate on the specific performance of sim2sim? And what do you speculate is the cause of its poor performance?

Thank you very much!

[BlackOtters]

GIFIn the simulation, by using the PD control of the g1_dof itself and the action scaling, the robot is unable to quickly and smoothly reach the target state and stabilize statically, unlike the hard_waist version. (num_training_step=200m)