Abstract: It is difficult for a robot to explore an empty hall environment, and the RRT (rapidly-exploring random tree) is difficult to expand rapidly in the environment with narrow entrances. In order to effectively solve these problems, a compound detection strategy of candidate target points is proposed by combining RRT and the frontier method. In addition, the cost value is taken as the evaluation criterion of the optimal candidate target point, and an effective cost calculation method is proposed. Furthermore, an improved TEB (timed elastic band) algorithm is designed to realize local path planning of the robot and ensure the robot to reach the target point smoothly. In the actual environment, three parameters of the proposed method, the exploration time, the driving distance and the exploration times, are 1 187.465 s, 97.551 m and 41, respectively; while in the simulation environment, they are 275.119 s, 130.051 m and 32, respectively, which are better than GTM (gridtopological map) and RRT methods under the same experimental conditions. Results show that the method can effectively solve both the problem of robot exploration in the empty hall environment, and the problem of RRT rapid expansion in the environment with narrow entrances.