Indoor mobile robot navigation through reinforcement learning (RL) represents a significant step forward in robotics, enabling adaptive, efficient, and autonomous navigation capabilities. Effective training of RL agents to navigate unseen environments requires diverse and interactable settings to generalize across various indoor layouts. While existing frameworks largely focus on generating images of home-like environments, this paper introduces a novel 2D environment generation framework designed specifically to create interactable office building layouts with diverse shapes, corridor structures, and spatial configurations. Additionally, we propose a partially observed map generation algorithm that is essential for running RL simulations effectively. Our framework provides training environments that foster the development of robust navigation skills suitable for complex, real-world office settings.