Self-training of bio-object (amoeba) locomotion system model which is operating by neuronet is considered. Amoeba model was prototyped by two ellipsoids connected with each other by thin channel. The length of the channel and the forms of ellipsoids are controlled by operating system. The motions of the model in the fluid are considered in Stocks approximation. Two scenarios of self-training are considered and their effectiveness are compared for different distances from target and initial object orientation angles.