Learning to Drive Using Sparse Imitation Reinforcement Learning