Balancing - Steering can also be done by hopping either end or both. Don't have to be moving much at all. Been doing that for decades. Obviously easier on the pedal bike.
Looking like we are rounding the basses. On the inverted pendulum side, we can certainly mostly agree. The more static side (bike) is classic case outside the fuel rolling around and the relatively minor lateral and vertical movements the steering induces. The wide tire effect is well documented on how to handle it. Interesting on degrees of freedom the test apparatuses were noted to have. On most that were mentioned, were only a few degrees before the control apparatus limits were exceeded. This may be significant with human control limits even with their ability to change a bit and help balance this all out.
On the dynamic side (rider), they can flex in multiple directions at the foot- ankle, knees, hips, waist, as well as flex the torso in addition to two arms and two legs that can expand- contract. Robot folks are still working this all out and come a long way. We have two pivot points to help balance, one under each leg that also can be torqued and articulated at the foot and joints above. Two legs possibly more like a very wide tire or more like two separate ones. Not sure how to simplify and model this all. Good exercise for certain. This all changes the combined COG as noted earlier.
It is looking much clearer to me at least why the bicycle trials guys all say No Way to clip in shoe cleats. Not using them allows much more foot placement latitude in addition to the easy foot off capabilities. Also helps explain the trend to wider foot pegs in the moto trials circles also. More like a shorter running board. either will allow some significant pivot point changes for them just for starters.