Integrating Vision-Language Models with Reinforcement Learning for Human-Aligned Decision-Making of Autonomous Vehicles