Pairs of magnets were installed in the path at certain interval and were detected by a magnetic ruler to correct accumulation error of position and orientation of inertial guided AGV. It was difficult to get the absolute position of magnets because of obstacles in job shop and this often caused trouble in debugging AGV. A teach-in method was used to record the position of magnets, and iterative learning control was introduced to update the aim point of AGV. After several times’ learning, AGV can run through magnets accurately. This method is easy to use in AGV path tracking. Accumulation of error can be eliminated when correction is done. The results of experiments in a tricycle AGV tugger were given to demonstrate the efficiency. 