Reinforcement Learning with Lie Group Orientations for Robotics