Evacuation guidance is a crucial response during disasters. Currently, humans provide this guidance, but they are at risk of being affected by secondary disasters. In recent years, the use of robots for evacuation guidance has been explored. However, there is limited research on methods that take stairway congestion into account. This study investigates an efficient evacuation guidance method using robots to avoid congestion. We propose an algorithm that specifically monitors congestion at stairways. Simulation experiments of a building fire showed that directing evacuees to less crowded stairs is the most effective approach, based on various evaluation metrics. The results highlight the effectiveness of the proposed algorithm.