Sean Klein
Andrew ID: smklein

PART 4: 
My "blind" can flipper is incredibly accurate. I would bet money on it being
able to flip a can in the range of 20 - 50 times without external human
intervention, as long as the initial can is placed correctly. 
This seems to be because the action of picking up/setting down the can 
"calibrates" the location, so to speak, to the whatever the robot assumes
it should be. So if the robot has gradual errors, these errors are also
added to the can. 

Adding vision actually makes it worse -- not because it is less accurate -- 
but rather, because the Kinect causes my code to crash before it can 
'misplace' the can. 


PART 5: 
The camera doesn't remain EXACTLY parallel to the ground (although it is 
kinda close), but that is partially because the joints on the 
Mantis' head are not symmetric -- looking 45 degrees left and looking 45
degrees right causes the Kinect to be slightly higher or slightly lower
on one side compared to the other.