The environment map is the basis of robot navigation. Autonomous robot exploration is the process by which a robot autonomously constructs a map in an unknown environment. How to make the robot reduce the traveling distance when completing an unknown environment is a worthwhile research problem. This work proposes a transformer-based decision network for autonomous exploration and a deep reinforcement learning framework for autonomous robot exploration tasks. Experiments show that our proposed method is feasible and saves an average of 6.7% on distance traveled compared to traditional methods.