I am working with openAI rllab and I would like to initialize the state of the mujoco simulation to the exact same state sometimes for debugging purposes. Is there a recommended approach for this?