A method, device and system of prediction of a state of an object in the environment using an action model of a neural network. In accordance with one aspect, a control system for a object comprises a processor, a plurality of sensors coupled to the processor for sensing a current state of the object and an environment in which the object is located, and a first neural network coupled to the processor. A plurality of predicted subsequent states of the object in the environment is obtained using an action model, a current state of the object in the environment and a plurality of actions. The action model maps a plurality of states of the object in the environment and a plurality of actions performed by the object for each state to predicted subsequent states of the object in the environment. An action that maximizes a value of a target is determined. The target is based at least on a reward for each of the predicted subsequent states. The determined action is performed.