In this research, tracking error compensation method using Neural Network is proposed in order to minimize tracking error for FRAGV. Dynamic model for FRAGV is developed for simulation, and simulations are performed for various cases. And application of the proposed method to real FRAGV system developed in MSD Lab at KAIST, is executed. Neural Network adopts multi-layer perceptron, and is learned by Gradient descent rule. In simulation, proposed method shows better tracking performance than conventional control method for complicated continuous path. it also shows better tracking performance than conventional control method for non-continuous paths such as tetragonal path and triangular path. In experiment, proposed method shows nearly equal tracking performance compared to conventional control method. It was proved by actual movement data using video recorder. From such a fact, possibility of application of the proposed method to real system is validated. There may be three possible reasons that tracking performance of the proposed method is not better than conventional method. The first is incorrectness of position estimation from Dead reckoning. And the second is insufficient learning because of limitation of experimental space. The third is worse experimental condition such as irregularity of floor. Complement for above possible reasons will give improved tracking performance than these results.