Human-Centered Reinforcement Learning for Lighting and Blind Control in Cognitive Buildings

In cognitive buildings (CBs), intelligent IoT edge devices do more than gather data. They also aggregate, analyze, and stream data at the edge of the network, where cognitive controllers based on machine learning algorithms enable new levels of control and security while significantly improving the overall user indoor comfort and safety. As a result, these CBs will be more intelligent, self-learning, innovative, and simple to manage. CBs also promise to heighten their dwellers' comfort (and, as a consequence, their performance) by optimizing, e.g., lighting, temperature, and humidity where needed. The reinforcement learning (RL) method is becoming more and more attractive for the designing of cognitive controllers. This chapter presents a human- centered reinforcement learning control for visual comfort management in cognitive buildings. A satisfaction-based visual comfort model is coupled with RL to adapt the boundaries of the comfort zone in the presence of a group of occupants. Compared with the traditional controls, it is personalized and human-centric since users' perceptions of the surroundings are exploited as the feedback loop. A case study of an office room and its performance are also presented.