The thrust will cause the car to topple to the left and this will free the roller _R2_, but a moment later it will bring the segment _G2_ in contact with roller _R2_ which thus receives an upward thrust. But an upward thrust, we recall, will not cause the spindle to move upward, but off to the right toward _C_; and so, a moment later still the roller _R2_ will pass beyond the end of the segment _G2_ and the roller _R1_ will come in contact with the segment _G3_, along which it will tend to roll, thus accelerating the precession to the right, and so causing the spindle to push downward, bringing the car back to its old position or beyond it; whereupon the segment _G4_ will be brought in contact with _R2_, retarding the further oscillation of the car and causing the spindle to move back again to the left. This sequence of oscillations will be repeated over and over so long as any disturbing force tends to throw the car out of equilibrium. In other words, the gyroscope, when its balance is disturbed by a thrust due to any unbalancing of the car, will begin to wabble and continue to wabble until it finds a position where it is no longer disturbed, and this new position will be attained only when the car as a whole is perfectly balanced again. In this new position of balance, the car (owing to a shift of its load or to the force of the wind) may be tipped far over to one side, as a man leans in carrying a weight on one shoulder, to get the centre of gravity over the rail, and in that event the axis of the gyroscope will be no longer horizontal. But that is quite immaterial. There is no more merit in the horizontal position than in any other, as regards the tendency to keep a fixed axis. If it is usually horizontal, this is only because under normal conditions the car will be balanced at its physical centre, just as ordinarily a man stands erect and does not lean to one side in walking. Reverting for a moment to our diagram and the explanation just given, it will be understood that the two rollers R1 and R2 are never in action at the same time, and that it is only the roller R1 that gives the sidewise push that accelerates the precession (since R2 is not in contact with the axle itself). The function of R2 is to retard the precession and bring the axis to its normal position at right angles to the rail on which the car runs. There is nothing of mystery about the action of either which the action of our gyrostat does not explain, but the mechanism by which the different segments of the car are made to push against the spindle, and so force it to balance the car in order that it may maintain its own balance, is exceedingly ingenious. Mr. Brennan himself tells me that he has improved methods of accomplishing these results, which are not yet to be made public. The principle, however, is the same as that outlined in the earlier patents which I have just described.