mismatched observations

[mateuszwyszynski]

It seems that step["gripper_open"] does not yield a correct value of the gripper state. It can be easily checked by comparing this value with images from the wrist camera. They mismatch significantly.

We currently use step["misc"]["joint_position_action"][-1] which aligns well with the images. There is no step["misc"]["joint_position_action"] for the first step, so the value is set to 1.0.

However, we should look into this.

• why the observations of the gripper state do not align with images? Are they based on some kind of sensor?
• can it be that there is an identical problem with observations of joint positions?