IM2HoT: Interactive Machine-Learning to improve the House of Things

Abstract

Patterns emerge as we go about our daily home lives. Can we take advantage of these patterns, learn them and use them to make people's lives more comfortable? Questions regarding a person's comfort in their own home, as well as security, healthcare, safety and energy savings drive smart home solutions development, with the purpose of making daily life easier, which can be achieved by remotely controlling devices or identifying relevant activities and increasing their automation. Current home automation solutions mostly focus on remote control, monitoring and preferences or scene settings set up by the user. Some of them, like the NEST Thermostat are capable of learning user patterns, but serve a single need. On the other hand, the autonomous control in solutions that allow for the integration of multiple devices is handled by the definition of rules, and cannot learn user patterns. The purpose of this thesis is the development of a solution that uses machine learning algorithms to learn user patterns, that integrates multiple kinds of devices heterogeneously to serve different needs and autonomously controls them, allowing for an off-the-shelf solution better suited to the users needs. Additionally, users will be able to define what devices they want the system to monitor and learn as well as help the learning process by reinforcing choices made by the system. This solution will be achieved by extending the visual programming environment NodeRed. The use of a visual programming environment will also enable users with little programming experience to easily interact with the system. There are currently no smart home solutions that allow the integration of machine learning algorithms to learn patterns and allow control over the devices that go beyond specific needs such as light or temperature. The integration of this system into NodeRed will also allow for a flexible and easy setup, potentially reaching thousands of users.