Smart Homes are environments that automate action and adapt themselves to user behaviours. In this sense, it is necessary to employ learning strategies to allow Smart Homes to truly become intelligent, in a sense that they anticipate needs and actions. This requires constant monitoring of environments, users and their actions, as well as, non-supervised dynamic learning strategies.

The purpose of this work is to develop a system capable of taking the best action possible based on its environment. In this document, we present a reinforcement learning approach to automate lights and appliances in a Smart Home environment. An intelligent agent perceives the ambient and the past interactions of the user with the home in order to learn what is the best action to perform, which action has a certain reward associated in order to inform the agent his behavior. A reinforcement learning algorithm learns a policy for picking actions by adjusting its weights through gradient descent using feedback from the environment.