We propose a disruptive, user-transparent occupancy sensor to accurately determine the presence of occupants in residential buildings to enable temperature setbacks and provide energy savings of over 30% per year. Our technique utilizes standard 802.11ac (or higher) commercially available WiFi-equipped devices such as routers, access points and Raspberry Pis (hotspots) as sensors to monitor an environment using the wireless channel state information (CSI) collected by these devices and occupancy-centric machine learning algorithms to determine occupancy from detected changes in CSI. Developed algorithms will be able to distinguish between humans and pets, to sense presence when people have been substantially still for extended periods of time, such as when they are sleeping, and to be robust to changes in the radio multipath environment owing to activities of daily living such as moving furniture or opening doors. The goal of the project is to demonstrate the novel machine learning algorithms that are key to the system performance and to optimize adjustable sensor system parameters so that a do-it-yourself (DIY) installable sensor system that meets the energy savings, costs and detection accuracy requirements of the SENSOR FOA can be demonstrated. The sensor system we propose is self-contained, beaconless, self-calibrating, capable of generating updated occupancy estimates in less than 10 seconds, and does not collect any personally identifiable information. The sensors and sensor hubs can be implemented using a wide variety of already commercially available devices that include standard 802.11ac (or higher) chipsets, including currently deployed routers, access points, thermostats and newer Internet of Things (IoT) devices such as home assistants (e.g. Google Home, Amazon Echo). While our sensor hardware components use so-called WiFi protocols to wirelessly probe an environment, they do not require nor utilize or require any access, WiFi or otherwise, to the internet or outside world. Therefore, we offer cost effective occupancy sensing to homes with and without cable or internet services or broadband access. The importance of our sensor hardware being based on so-called WiFi devices is that we can take advantage of components and devices that have already been pushed down the manufacturing/learning cost curve. We have assembled a uniquely qualified team to perform the program tasks and to transition the technology to commercial viability. The program will be led by Principal Investigator (PI) Mr. Eric Giler of Endeveo Inc., a serial entrepreneur who successfully commercialized the first voice mail systems, mobile music platforms and mid-range wireless power delivery systems. Mr. Edward Vineyard, group leader at Oak Ridge National Laboratory and an active member and Fellow of ASHRAE, and recipient of the ASHRAE Exceptional Service Award will lead the energy savings study and machine learning algorithm simulation and validation. Professor Anthony Rowe from Carnegie Mellon University, an expert in next generation sensing systems and networks will lead the algorithm development activities as well as the WiFi platform energy and computing resource optimization activities.
