Physical-Intelligence
mobile aloha gripper is not moving properly
with the code you shared, I fine-tuned and tested it, and found that the arm follows the intended trajectory well, but the gripper repeatedly opens and closes.
we have modified the aloha gripper(hw change)to have the following settings PUPPET_GRIPPER_POSITION_OPEN = 0.05800 PUPPET_GRIPPER_POSITION_CLOSE = 0.0449
PUPPET_GRIPPER_JOINT_OPEN = 2.487 PUPPET_GRIPPER_JOINT_CLOSE = 1.5
It seems that we need to modify the aloha_policy. adapt_to_pi = True after setting can I fix the following three functions?
def _gripper_from_angular(value) value = _unnormalize(value, min_val=0.4, max_val=1.5) return _normalize(value, min_val=-0.6213, max_val=1.4910) # fix: value = _unnormalize(value, min_val=1.5, max_val=2.487)
def _gripper_to_angular(value): value = _unnormalize(value, min_val=0.01844, max_val=0.05800)# fix: value = _unnormalize(value, min_val=0.0449, max_val=0.05800)
def _gripper_from_angular_inv(value): value = _unnormalize(value, min_val=-0.6213, max_val=1.4910) # fix: value = _unnormalize(value, min_val=1.5, max_val=2.487) return _normalize(value, min_val=0.4, max_val=1.5)
Hey @a510721! If the gripper exhibits jumpy behavior it's probably because of a bug in the scaling of the observation and action. Based on the values you shared, it looks like your Aloha does not match our internal Alohas or the external Alohas we tested on. Your changes generally look to be in the right direction although in _gripper_from_angular_inv make sure the change applies to _normalize and not _unnormalize. Your max_val for _gripper_to_angular also looks a bit small given a min position of 0.0449. I would double check those values are correct.
