Randomness is one of the familiar concepts in our daily lives. When what result will be obtained is not clear in advance, we often predict the result using probability. Once the result is obtained, we may regard it as one of the possible futures that was selected as a sample and became reality. Taking account of which events may occur with what probability is important in making better decisions. To achieve automatic control with this philosophy, we need to deal with stochastic systems in which the information on the randomness is reflected. In this talk, I will first introduce a stability theory that lays the foundation of control for such stochastic systems. In the theory, the stochastic processes determining the system dynamics are not required to be time-homogeneous nor stationary. By exploiting this flexibility, I will propose control of stochastic systems that is adaptive to the temporal variations of the probability distributions behind the systems. Then, I will also introduce our ongoing study on the model predictive path integral (MPPI) control, a kind of sampling-based model predictive control that could be directly integrated with ensemble data assimilation, and conclude this talk with showing my interest in the integration of control with data assimilation.