Thanks for the quick reply!
As far as I understand stall detection in 0.5.0 runs in a different thread, and has a cycle of ~4ms.
In my (admittedly limited and potentially faulty tests) stall detection didn't work very well at "high resolution" (e.g.: 10ms).
I wanted to experiment with initial calibration done at high (10-50ms) resolution, perhaps completely bypassing the regulator code to get the current reading.
If I can make this work fast & safe enough (not overheating the motor) - this could be useful beyond initial calibration.
In my construction there are two potential "stall-like" conditions - one is grip trying something too heavy (lesser drift between reading and actual tacho count), and second one is hard stop (currently done using touch sensor).
Hard stop is used for calibration, and delta between tacho reading on hard-stop during calibration and operation is used to determine if "grab" has been succesful.
BTW: some pictures if anyone's interested, grip is towards the bottom of the page http://www.fuzzycow.org/fuzzybots/holon ... w/gallery1