Describir: Deep imitation reinforcement learning for self‐driving by vision