In this paper a reinforcement learning based nonlinear control strategy for control of boost converters is presented. Control of boost converters is a challenging nonlinear control problem, and classical linear control techniques perform poorly since the model of the converter depends on the state of the switching elements. In this paper the boost converter control problem is formulated as an optimal multi-step decision problem aimed at attaining a constant output voltage. Optimal multi-step decision problems can be solved using the framework of Markov Decision Processes (MDP) and Reinforcement Learning (RL); however iterative solution procedures exist only for discrete state problems. In this paper two possible approaches for applying RL to the boost converter problem are proposed. First a RL based control strategy for a discretized model of the boost converter problem is presented. Next an approach that applies robust regression to mitigate the effects of discretization by smoothly interpolating between the control decisions computed for the discretized states is presented. Simulation results indicate that the robust regression based RL strategy significantly reduces oscillations and overshoot and gives a better output voltage compared to the pure RL strategy. © 2016 Published by Elsevier Ltd.