There is a size of push no fixed stance can absorb, and then balance stops being about stiffness and becomes about where you put your next foot. Roboticists reason about this with a deliberately simple model: treat the whole body as a point mass on a massless leg, a linear inverted pendulum. While one foot is planted at position p, the center of mass accelerates away from it as omega squared times its offset, where omega is set by the pendulum's height. That gives a conserved quantity, the orbital energy: velocity squared minus omega squared times offset squared. If it is positive the mass runs away and you fall; if it is exactly zero the mass glides to a stop right over the foot. So recovery is a placement problem. Solve orbital energy equals zero for where the foot must land and you get the capture point: the current position plus the velocity divided by omega, a step out ahead in the direction you are falling. Step there and the fall is caught. This is the idea (Pratt's capture point, Kajita's inverted-pendulum walking) under real stepping controllers; the full-physics version, stepping every cycle on a real biped, is what the pilot and Forge stacks take on.