It does indeed work! I've been meaning to instrument a food stimulant mass to determine how the chaos is effected my mass offset vs speed. I think due to the under-actuated nature of the system you're stuck with a balance between cg-offset, gravity and input rotation on the first axis