Optimizing Bipedal Maneuvers of Single Rigid-Body Models for Reinforcement Learning