In this paper, an offline algorithm generates gaits for robotic biped running motion. A minimization procedure on a novel objective function produces initial joint velocities for the flight phase given the initial joint angles. These computed initial velocities ensure a reasonable starting point for the constrained dynamic optimization procedure to try and formulate a gait that minimizes energy consumption. The simulated robot runs even on ground and up stairs. Results demonstrate the rigid robot solution has a intuitively satisfying human-like motion with an energy consumption that appears reasonable for a rigid robot.