The invention provides a robot localization method based on self-adaptive Monte Carlo localization method. The method comprises the following steps: prefetching cartographic information of an environment in which a robot is arranged by adopting a prefetching technology; calculating similar potential energy areas in the movement process of a to-be-localized robot; localizing the robot by adopting a Monte Carlo localization method of a self-adaptive sample according to three-dimensional grids and similar potential energy areas. The method not only has the advantages of the conventional Monte Carlo localization method, but also can solve the localization problems of position tracking, global localization and robot kidnapping while the diversity of the sample and high efficiency of the calculation can be guaranteed. The beneficial effect of calculating the potential energy and values of all sensors is that the direction of the robot does not need to be considered, thus one dimension can be reduced, and the calculation amount is greatly reduced.