Autonomous buses navigating pedestrian zones face challenges due to unpredictable pedestrian behavior and dynamic obstacles. While previous research on autonomous bus navigation has demonstrated success in controlled environments, existing control and planning approaches often struggle to adapt to the complex, unpredictable nature of pedestrian zones. This paper presents a reinforcement learning (RL) based end-to-end navigation approach designed for autonomous buses in pedestrian-rich zones. The RL model was developed and tested in a high-fidelity simulation of the RPTU campus, created with Unreal Engine. Utilizing minimal sensor data (GPS and IMU), the agent achieved goal-directed navigation within a static environment at a capped speed of 6 km/h, converging in approximately 20 min of training.