Deep Q-Learning (DQL) has shown promising results in autonomous navigation tasks, yet achieving consistent path following while avoiding obstacles remains challenging. This paper presents a DQL-based approach for autonomous vehicle navigation that effectively combines path following with obstacle avoidance across multiple environment configurations. While methods like PID-PSO fuzzy controllers typically treat path following and obstacle avoidance separately, recent approaches such as RL Coach and deep learning-based trajectory planning have aimed to integrate these objectives to some extent. However, these methods either require significant computational resources or are limited to structured environments. In contrast, the proposed approach integrates both objectives seamlessly into a unified learning framework, leading efficient navigation behaviours. A neural network setup with a dual-network system is used to keep learning steady. The agent tries out different things and remembers past experiences to learn better and make smarter choices. It works well on different paths with static obstacles. The agent figures out direction to move around these areas, sticks to set routes, and stays away from obstacles. The proposed method has been evaluated by observing the Mean Squares Error (MSE ~0.03), the reward system, and number of steps per episode. It avoids obstacles about 81% of the time, suitable for real world application. Computational efficiency is examined to ensure hardware suitability of the proposed work.