New RL technique achieves superior performance in control tasks