The invention discloses a method for controlling a mobile robot through automatic establishment of abstract action. The method comprises the steps of 1, initialization; 2, updating all state-action assessed values according to a Q-learning module updating formula, saving an experiment track, and removing the states in repeated state intervals in the experiment track; 3, calculating the diversity density values of residual states; 4, executing step 5 if the diversity density value of a certain state reaches a preset threshold value, and repeating step 2, 3 and 4 otherwise; 5, taking states with diversity density values larger than the threshold value as sub-goal points, establishing an abstract action module, setting the start state, strategy and stop condition of the abstract action, adding the abstract action module to a selectable action set of the robot, and controlling the mobile robot with the updated selectable action set. By the adoption of the method, the abstract action can be effectively established, the robot can conduct hierarchical reinforcement learning by means of the established abstract action, and then the learning speed of the robot in a complicated environment is increased.