A method of generating an optimum parking path of an unmanned driving vehicle which is performed by a controller in the unmanned driving vehicle, wherein the controller changes a moving distance for a plurality of operations in a reference parking path, finds a parking path in which an average obstacle distance, which is an average distance between at least one near-to-path obstacle and the unmanned driving vehicle, is longest among a plurality of candidate parking paths, and sets the parking path having the longest average obstacle distance, as an optimum parking path in response to the longest average obstacle distance being longer than a predetermined limited distance.