Active exploration plays a significant role in improving robotic autonomy. It enables robots to acquire information about the environment without human intervention. While single-agent active exploration has garnered significant attention in recent years, multi-agent active exploration has increasingly attracted interest due to its efficiency. In this paper, we propose a multi-agent active exploration framework that not only inherits the efficiency of its predecessor but also is compatible with multi-agent working settings. Our framework consists of three key components: i) a door information-based local topology map generation module, ii) a global topology map fusion module that fuses the local topological map into a global one, and iii) a policy module that assigns goal points for each robot and converts them into action instructions. We demonstrate the efficiency through experiments on the Habitat simulator. Experiments on real-world robots are also implemented to demonstrate the possibility of transferring to real-world environments. A video demonstration of our algorithm’s performance is available at google drive .