For typical robot locomotion like walking on even terrain, a limited prediction horizon of approximately 1-1.5 seconds is sufficient, with the control system recalculating at roughly 10 times per second, updating predictions based on current obstacles and foot placement.