I also plan on using geared DC motors with encoders to do closed loop control as one of the things I'm trying to do is build a really quiet printer and I think that should be easier to achieve than if I used steppers.
Us and downs, but I'm feeling optimistic at the moment. This might just work!
I AM looking at the Wally and I have tons of ideas for calibration. I kinda got to this idea via wanting to build something with DC motors, eventually realised that it is very similar to the Wally, so started having a closer look at it. If everything really fails I'll go back to something cartesian because building a quiet printer is most important to me.
i am currently building a gus and in spite of its simpler mathematics it is a pain to calibrate. i think wally's kinematics is even more complicated and that makes it hard to calibrate. I think another aproach could be to use two arms of gus horizontally and move the z platform. that will be easy to calibrate. just a thought. i am not sopping u from your en devours just warning you of possible pitfalls and time wasted fixing them.
I figured with all the money saved on precision linear components I'll just blow all the savings on some crazy ideas for calibration. I thought that once you have it calibrated, something like this will really help with storing the calibrated position: [ams.com]
Got some samples a while back and it is simple enough to work with. It says 14-bit resolution in the summary, but I think I remember from the datasheets that you only really get about 13-bits of resolution. You can take lots of measurements and average them out which seems to help a lot too which is OK to do when stationary at the start of a print I think. This type of sensor was one of the main reasons why I got interested in SCARA or polar type bots over cartesian ones in the first place and it is kinda why I want to at least try this out first.
I'm also happy to throw in redundant calibration methods like opto-interrupters or limit switches, manually dragging the arms out to form a perfect equilateral triangle with a jig to line it up, small test prints or plots with trial and error until you form exactly the desired shapes, etc and then just average out to reduce the errors or even fudge values until you get it right.
And I'm more than happy to base things off a more powerful ARM based board to help with the increased complicated calculations.
Clearly I'm still in the hopeful, ambitious phase of this project