The emergence of autonomous driving technology has sparked interest in the concept of end-to-end autonomous driving. This study investigates end-to-end autonomous driving using a deep reinforcement learning approach based on the double deep Q-network (double DQN). A control model is developed for end-to-end autonomous driving using both the DQN algorithm and the double DQN algorithm. This model utilizes RGB images from the vehicle’s front camera as input to directly determine the steering angle control. Training and validation are carried out in the AirSim simulation environment. Results indicate that the DQN algorithm achieves convergence after 50 training episodes, while the double DQN algorithm reaches convergence in just 20 training episodes, demonstrating faster convergence. Additionally, the DQN algorithm shows fluctuating loss curves in later training stages, whereas the double DQN algorithm maintains consistent performance and stability throughout training. These findings suggest that the double DQN algorithm offers improved convergence speed and stability compared to the DQN algorithm.