To return or not to return... and I don't really know Python

What kind of precision you need for those operations. If you could shoot a video with doing by hand what you need a robot to do, that would really make figuring out your needs faster and much more accurately.

If you are planning to make a robot to take a sheet of paper from a pack that probably would need some advanced end effector (like in a printer) in the first place because it is not an easy task to take only one sheet, and such an advanced (hypothetical) end effector may weigh too much for Dobot to lift.

Another problem is for the robot to find the configuration of the objects (with a DRGB camera?) to figure out how to take them. That area is still actively researched by brightest minds and has been for many years.

Only based on the configuration of the objects you need to manipulate (their possible location, orientation, proximity to each other, likelihood of being displaced and many other conditions) one can say whether a specific machine would be able to handle the task right away, or additional tools/end effectors/mechanisms may be needed, a research may be needed, or whether it makes any sense at all to try to use that machine for that task. That is why a video would help a lot.

Well, you are saying that this video is not exactly what you want it to do, in which case I can't guess and can talk only about that video.

The robot on the video is not poor quality, it is the choice of motors that makes the robot move this way. Servo motors, as opposed to stepper motors, do work this way, especially cheap ones. The motors inside those cheap servos are just DC motors that are driven by feedback from the electronics reading an analog signal from a potentiometer. There are much more complex servos (e.g. with high resolution encoders inside and fast and complex electronics) that behave much better in these tasks, but the bottom line is that they are closed-loop systems and their moves are based on their position at every moment. The cheapest variant stops the motor and returns it only after it overshoots the position you tell it to go to. The position is sent via PWM in this case. You can see the same behavior on the servos on dobot that drive tool rotation and gripper. Cosmicray made a video of the phone panning with dobot that shows that that was never going to work. See youself. The link is somewhere on the dobot's forum. Or try is yourself.

But back to your task. That robot on the video (let's call it atmbot) might actually be doing that particular task better than dobot because atmbot uses servos which are closed-loop systems and there can no slipping occur. Slipping, in case you don't know, is when a stepper motor skips steps. If slipping occurs (e.g. when dobot hits an obstacle) dobot thinks it moved to the specified coordinate when it actually didn't and it is no longer in sync with the world and expect it to do things in a virtual world which coordinate system is skewed. So it depends on how hard you have to press the keys on that atm. If hard enough at arms' angles enough for steppers to slip then you're screwed. Servos don't have slipping and will just push harder up to their limit to get to the angle you set with PWM. Also, cheap servos always move with maximum velocity and by the time atmbot hits a key it may have enough kinetic energy to actually press it. You can't rely on kinetic energy with dobot because it will most likely slip.

Having that said, even programming atmbot is much easier that dobot for that task because it is very forgiving about hitting something. Dobot has no sensors when it operates as opposed to atmbot reading potentiometers inside servos at all times. Dobot reads accelerometers only once, when you power it up (or start dobot app) to estimate the end effector starting pose. After that dobot just keeps track of how many steps each motor has made.