DARPA’s Learning for Locomotion project sponsors teams throughout the nation to engineer competitive generalizable solutions to difficult quadrupedal locomotion problems. These systems are implemented on the LittleDog robot platform designed by Boston Dynamics and are tested monthly at a third-party government facility. The CMU team has successfully developed winning solutions to these tests by decomposing the problem of quadrupedal locomotion into two stages. The first stage plans an optimal sequence of footsteps while the second stage computes a feasible trajectory through the robot’s configuration space to execute those footsteps. This footstep planning approach has previously proven effective in complex bipedal locomotion problems as well (Chestnutt et al., 2005). Two applications of my imitation learning algorithms to footstep planning are linked below.