A practical motivation for developing legged systems is to enable agile locomotion in environments with variable terrain and limited foothold availability. A variety of challenges also make legged system control a rich field for academic study. For example, legged systems use intermittent contacts to generate external forces; there is potential to exploit passive dynamics and internal degrees of freedom; mobile systems are subject to noisy information and time-varying environments; and increasing the dimensionality of such systems produces the complementary effects of potentially improving performance while simultaneously making analysis of such “improvements” inherently more challenging to quantify. In this talk, we discuss work at UCSB on one-legged (hopping), two-legged (biped walking) and four-legged (RoboSimian) robot models. Each system facilitates its own contribution toward agile and robust legged mobility. We highlight work on reachability-based control for underactuated hopping, quantification and optimization of robustness of biped walking, and various techniques for planning dexterous and dynamic motions with RoboSimian, a dexterous, four-limbed robot developed by JPL for the DARPA Robotics Challenge.