I did a couple tests of SAS in 0.21 to see how it performs during rocket launch. Here's my results with videos. Test 1: Normal launch of the craft. SAS is enabled from launch. There are 6 reaction wheels, one at the top of each booster under the nose cones. In this test SAS is able to keep the craft on course. I keep the last set of boosters in place to destabilize the craft somewhat in the roll dimension. There is no input from me other than the gravity turn. Note that SAS is using some Roll correction to keep the craft stable. http://youtu.be/JXINZ7P6W6k Test 2: This didn't go as intended. It's similar to Test 1 except the booster separation stage goes poorly. When the second stage of boosters disconnects, it damages a remaining booster sending the craft into a spin and ultimately causing it to break up. I let the flight continue to see if SAS would correct the spin. Initially it does. However after the spin is damped, it continues to applies a massive over correction reversing the spin. I've developed simulators similar to this (ground vehicles) before, also using PID controllers to control HPR. I'm wondering if the game tracks rotation beyond 360 degrees or 2pi and is actually trying to unwind the craft through all the rotations that have occurred? http://youtu.be/eYFKu1xVdxk Test 3: This is very similar to Test 2 except I intentionally transfer fuel to a burned out booster to introduce spin. Again the SAS does correct the rotation initially, but continues full force over correction after the spin is damped. Toward the end I use RCS to stabilize the craft with SAS still enabled. Again I'm wondering if it's trying to unwind all the rotations that have occurred instead of moding the crafts HPR angle by 360 or 2pi. http://youtu.be/A-Nrt8zka0A The spin is introduced at 2:50.