And today, on your preferred Mystery show: The Progressive Error.
Me (jeez) and Filipe (calega) have been struggling around with the motors calibration on the past few days while the girls have been busy preparing the prototype to be presented in London.
It has been quite a rush here, but thins are going on. The main issue now (and probably, the last one) is that the Etch-a-Sketch (TM) appears to have an intrinsic error due to its mechanical design. Just grab one of these toys and try them. You can see that when changing directions you get a play error on each knob. In other words, you have to “waste” a few degrees off before actually see something being drawn again.
Next step is to figure it out a way of compensate this error and we are already trying to measure it and raise a software solution. But check the pictures below… they are kitsch art, right?! 
Our first circle: 
A circle after some quick motors adjustments:
The famous beach picture “Coco Gelado” (which means “Cold Coconut”):
A N900:
The PUSH contest logo:
ok, PUSH fellows, I own you a better drawing of your logo!
jeez
Ohh, I forgot to answer the general question:
I have fixed a resolution of 150 dpi on this, so we are getting more than 800×600 on this Etch-a-Sketch and more than 1024×768 on Filipe’s one.
by jeez on January 28, 2010 at 3:31 pm
I imagined you were going to face something like that… I believe I even commented previously. That thing is just too imperfect, and should add lots of noise to your track. To make perfect drawings you would need some kind of feedback. From the camera, for example, in a remote future. An estimate of the actual motor motion from a wheel encoder might help too. The problem you are facing is just like the problem of localization in mobile robotics, where you can’t trust your estimate of the position because of the wheel slippage, and make is necessary to have some kind of position sensor.
You might be able to model this error somehow, and try to deal with it in software. But what I would suggest, considering you don’t have much time available, is just make sure the drawings are made following just one general direction, with few loop closures. That way you will minimize the distortions created by the noise…
But that is all more curiosity and pedantry than anything. The examples so far look very, very nice!!… Well done!
by nic on January 28, 2010 at 4:41 pm
I know you’re still working on it, but computer’s interpreting pictures has always been a special kind of voodoo to me, something I just don’t understand well. I’m impressed you can get ANYTHING to work, let alone as well as you have. Bravo!
by Mike77 on January 28, 2010 at 9:04 pm
Nice to hear this, guys!
Thanks for your tips and good words!
Best Regards!
by jeez on January 29, 2010 at 5:34 pm