Mobile robot navigation can be stated into following problems: detection of obstacles, finding a free path leading from the robot's position to the goal position, estimation of the robot position and realization of the path. In our experiments the mobile robot B-14 is used. This vehicle is equipped with 16 sonars, 16 infrared sensors, on-board Pentium computer and one colour camera. A grid-based map of an environment is built on the base of sonars and infrared sensors.The status of each cell of the map can be free, occupied or unknown. The map is built up incrementally and fuzzy method of data aggregation is used. The mapping method requires knowledge of position of the robot within its environment. Therefore a self-localization module is necessary. In presented navigation system artificial visual landmarks are used in order to compute the robot's position. We show experimentally that the resulting method is able to localize efficiently a mobile robot in partially known environment.

Path planning problem is defined as follows: knowing the robot position and the goal position find the continuos and collision-free path leading from the goal to the robot. We used diffusion method for path planning. This method was proposed by Steels in 1988 and implemented by Siemiatkowska in 1994 in a form of two-layer cellular neural network. This implementation allows on-line path planning. The method has following advantages: there is no problem with local minima, situations when the obstacles surround the robot or the goal are easily recognized [7]. We also used CNN for image processing [12] and self-localization [9].

New method of self-positioning of an indoor robot equipped with a laser range finder was developed [1][13]. D-histogram is a new method of processing two-dimensional range scans obtained using a rotating laser range finder. The method searched for a segment of collinear readouts by the way of finding groups of neighboring data that belong to the neighboring lines bearing the same orientation. The extracted segments may be used in a search for predominant wall directions in a scene. This information is very useful in correcting dead reckoned orientation of a robot working in structured or partially structured environment. The method is inexpensive and robust against noise.

Mapping is a process of building a model of an environment based on sensors measurements. The environment is usually represented as a topological or a metric map. A topological map is usually represented as a graph. Metric map can be geometrical or grid-based. Grid based representation was proposed by Elfes and Moravec. The robot's environment is represented as a grid of cells, and an occupancy value is attached to each cell. This value represents our belief that an obstacle occupies the corresponding region. When new value of sensors measurement are obtained the map of environment is updated - possibility values computed based on sensors indication are combined with corresponding values of the global map.
Algorithm for aggregation information taken from different robot's position and different sensors were implemented. Following methods of aggregation has been tested - probabilistic approach, fuzzy logic and Shaffer-Dempster theory [3][6][9].