(...)
At the moment, I just start autoguiding, take a shot, manually stop auto guiding, move slightly the mount, so the star on the screen of the guider moves from one side of the frame to the opposite side, start guiding again, and take another shot, and so on. Sometimes I don't need to even manually hit buttons, since when I stop guiding the star starts to oscillate around (maybe because of periodic error).
I see, you do manual dithering. It's nice

But don't forget what if you want a good dithering, displacement have to be 'randomized' in both two dimensions and not only in the RA axe (because PE).
It would be wonderful to automate dithering, but how? Maybe I could just start guiding and acquisition sequence on the camera, and set a timer somehow to stop autoguiding every, say, 6 minutes, during the pauses of the acquisition (commands GUIDE -> STOP and GUIDE -> RESUME).
With AG (auto-guiding) the scenario should be:
1 start >> 2 AG is running >> 3 start shoot >> 4 stop shoot >> 5 stop AG >> 6 random displacement >> 7 start AG >> 8 AG stabilized >> GoTo 3
It needs communication to sync all events because sometimes step 6 or 8 takes 10s, but sometimes it takes 30s. Plus, set intervals in different devices is hard to be sync after 1 or 2h runing; it could be out of sync for 1 or 2s so you ruin your shoot.
Interesting. Can you give me more details about it, or a link to read?
(...)
For AG + dithering with a laptop you can find a couple of software what can drive mount, imaging camera, guiding camera and more (focuser, etc.) via a standard protocol: ASCOM. With that you can automate all your night.
Example, with a light mobile setup for wild field you can use software like
PHD Guiding or
Guidemaster after installed the
ASCOM platform. Many astronomers use this kind of setup.
Sorry to not be prolix but I'm not sure what it's the right place here to detail all parts.
(...)
Personally I think it's better to spend as much time as possible "collecting photons" rather than stopping and waiting for the camera to cool or taking darks, and then just making sure I get darks later for all the various temps my camera was at and properly calibrate my frames. Usually the next day after a night of shooting I look at the temperature in the exif data and shoot darks for every unique temp and shutter length that I see. I get my camera to the temps it was at by putting it in the fridge to cool it down or running liveview to warm it up. This way I don't waste any time I could have spent exposing (or sleeping). But if you have some evidence that giving the camera a break to cool is actually better, I'd certainly amend my current process.
I'm totally agree with you: signal is the most important.
But where signal is, noise is. So we need to process images with darks, bias, etc. if we want to get 'only' the signal ('only' is a dream).
Make a dark library is very interesting, specially for 600 or 1200s shoots (in some case night is too short to get lights + darks).
But (I'm sorry to say 'but' all times), its very hard to get 'similar' darks (you can do the test with two darks at the same temp, one from 3 months ago and one just right now). The only thing what is a bite 'stable' is bias.
About temp rising during shoots, it's important to use the dithering time to cool the CMOS, specially with big 24x36 CMOS.
Take a look in this interesting page, on the bottom 
It's possible for ML to send out signals through various means. You can send out data over the USB with PTP, also for something more simple, turning on/off the audio preamp causes a detectable signal in the audio line in. Don't know if that would be helpful for your dithering. I cannot afford an autoguide setup, so my cheap eq mount's periodic error does this for me :p The draw back is that I could never hope to get 600s like you mention, 120s is my practical limit and that is usually only for small declinations, and if I spend a lot of time getting the polar alignment perfect.
It would be amazing to use ML to drive dithering, bulb shoots, etc. !

I think it's possible (if we could send orders on ST4 guiding protocol etc.), but wahoo... It's a big project.

(...)
(
http://starshoot.over-blog.com/article-le-gso-rc-8-sort-du-bois-71292056.html
http://jcbphoto.free.fr/astro/M51_110411_version_01.jpg with my RC and NEQ6
)
(...)
Woho, I see we are some French astronomers here

Edit: corrected links