This thesis considers the control of clutches on the Double Pendulum Robot which allows the introduction of passivity into actuated joints. Considering the task of throwing using minimum energy, the goal is to throw a ball to a minimum desired distance. As this problem involves active and passive actuation, as well as inherent discontinuities due to the clutches, the control problem becomes difficult. However, an optimization approach for the underactuated throw is proposed that is able to cope with the latter mentioned difficulties. In the process, discretization as well as dynamic programming incorporating Value Iteration is utilized to find the solution. The method is not restricted to throwing and can also be applied to other tasks. In order to discover the full extent of passivity in the double pendulum, unactuated and underactuated throwing are exploited. For the unactuated throw, advantageous initial conditions that lead to long distance throws are found. Moreover, throwing trajectories using underactuation to improve energy consumption are presented and evaluated in simulation and experiments.