In theory, using spherical cows, sure. But in practice there are all sorts of things screwing up the "probability" when jumping. Slope of the ground, windspeed, ground cover, etc.
Then it's still dependent on some known inputs: final position depends on initial position, wind speed/direction, ground conditions and gradient, etc etc. There is randomness, but you could make a probability distribution.
Ok? But it won't just converge to 0,0 because everything isn't equal and opposite, there will be some sort of bias that will move the bean away from it's origin.