The best result shown by sensor fusion (odometer and gyroscope). Gyroscope by himselft isn't too precise. To measure with him I took average speed of both wheels and angle, assume that robot moves in that direction. Nevertheless it works. I can not apply Kalman's filter to results of this estimation but I did apply to sensor's data directly. Only after that I handle x and y values.

Better results shown by odometer, it has more accuracy. Also, I can't apply Kalman's filter to results of measurements. Because of its non-linereaty.

Woth to note, in all cases I used a magic_coefficient to transform odometer's data into velocity. It equals 0.055 which near to the 'real' value 2*pi*2.7/360~0.047