For navigation of a mobile robot, the determination of its absolute location is one of essential tasks. Recently multi-sensor system is prevalently used for the accurate estimation of the robot``s position. We adopted the vision system with a cleverly designed landmark as an absolute sensor, and a dead reckoning system improved by the neural network as a relative sensor. Vision system is generally inappropriate for real-time applications due to the large amount of computations. In order to overcome such difficulty we design a landmark of an efficient pattern and develop a fast algorithm to calculate the distance and orientation of a camera with respect to the landmark. Exact and closed form solution of the camera location is obtained from the geometric relation between the pattern of landmark and its projected image. Moreover the presented algorithm requires only one row of mark image. Thus determination of the location can be processed in real time. In addition, the mark can be identified using projective invariant of the original mark pattern without any additional patterns. The effect of the error sources on the location determination is analyzed through a series of simulations and experiments. The applicability of the algorithm to mobile robots is discussed using the experimental results. On the other hand, the conventional dead reckoner is liable to give us wrong positional information especially when the wheel slips, because it depends merely on the angular velocities of the wheels. Accordingly the linear velocity simply converted from the angular velocity should be compensated. We propose the neural net based compensator to estimate the linear velocity of each wheel and to detect the wheel slipping. In order to determine the structure and input variables of the neural net, we analyze dynamic characteristics of the wheel slipping. After training the networks for various conditions, we test to detect wheel slipping and to estimate the linear velocity and t...