Funded by NSF NetSE Program (NSF-0914371, $300,000, 10/2009-9/2012)
Information and communications technologies are poised to transform our access to and participation in our own health and well-being. The complexity of this challenge is being shaped by concomitant transformations to the fundamental nature of what it means to be healthy. Having good health increasingly means managing our long-term care rather than sporadic treatment of acute conditions; it places greater emphasis on the management of wellness rather than healing illness; it acknowledges the role of home, family, and community as significant contributors to individual health and wellbeing as well as the changing demographics of an increasingly aging population; and it recognizes the technical feasibility of diagnosis, treatment, and care based on an individual’s genetic makeup and lifestyle. The substrate of 21st century healthcare will be computing and networking concepts and technologies whose transformative potential is tempered by unresolved core challenges in designing and optimizing them for applicability in this domain. (Source: NSF)
A smart environment contains many highly interactive and embedded devices as well as the ability to control these devices automatically in order to meet the demands of the environment. While smart environments offer many societal benefits, they also introduce new and complex challenges for wireless network design. A typical home may be equipped with hundreds or thousands of wireless sensors that aid in ensuring the health, safety, and productivity of its residents. If these sensors are continuously operating in full-alert mode, they will expend a great deal of energy and bandwidth. The result is an expensive infrastructure that requires constant maintenance to replace batteries and ensure quality-of-service. The goal of this project is to imbue such wireless sensor networks with cognitive capabilities and context awareness that will allow them to act in a more intelligent manner. The principal investigators (PIs) will use machine learning techniques to recognize activities that are being performed in the smart environment. This context information will then be conveyed to the network to allow sensor nodes to intelligently decide when to sleep, when to wake up, and how to route information. By transforming sensor networks into activity-aware sensor networks, the researchers hypothesize that they will greatly reduce energy and bandwidth consumption.
The contributions of this project include: 1) enhance existing algorithms to recognize activities that are incomplete, interleaved, or performed in parallel by multiple residents, 2) design and implement an algorithm that will allow each sensor to intelligently decide sampling rates and sleep/wake times, and 3) test the algorithms in simulation and in two physical smart environment testbeds. All of the synthetic and real-world datasets will be disseminated, together with the source code, over the Internet to facilitate community-wide comparison and collaboration.
Broader Impact: The research enhances pervasive systems with cognitive capabilities and context awareness. By partnering sensor networks with intelligent reasoning and learning capabilities, The PIs are proposing a paradigm that can be used to create innovative intelligent sensor networks, sustain smart environments, and improve the quality of life for residents of smart environments.
Creating a sensor network for a smart home poses new challenges and opportunities to the networking community. The setting for this application is different from many existing sensor network applications, as it is neither low-duty cycle nor high-fidelity environment monitoring. Instead, the sensing, transceiving and networking shall be driven by the home activities to strive for both long network lifetime and 24×7 continuous monitoring. For easy of installation at home, sensor nodes are powered by small batteries only, hence the network must minimize energy consumption to extend the network lifetime (e.g, no less than one year to be acceptable for consumer use). At the same time, it must provide 24×7 high-quality continuous monitoring, as the consequence of unreliable homecare may be life threatening, and can not be tolerated by consumers. On the other hand, this application provides an opportunity to study activity-aware sensor networking, with the integration of activity recognition algorithms described earlier. For instance, if no one is at home, the sensing and networking activity shall be close to zero in order to reduce energy consumption. If resident is at home, the sensing and networking needs to provide reliable and predictable monitoring. PI Cook has pioneered research on smart environment technology with low-cost sensors, and PI Song has leading research and development experiences on wireless sensor networks. The smart environment technologies will provide information that adds intelligence to the wireless sensor networks. Imbued with cognitive capabilities and context-awareness, we offer a new paradigm in which a smart home sensor network may seamlessly embed relevant context information into the network architecture, protocols and services. Our proposed ActiSen system architecture is illustrated in the following figure:
When ActiSen is running in Monitoring mode, the sensor nodes will detect events and perform activity recognition analysis. Eventually, these nodes may be able to infer additional contextual information. Because of the sequential order in which sensor events are generated within activity models, sensors will learn which other nodes are neighbors within the spatial configuration of the environment (these are typically nodes which are found in the same activity sensor set). Also, because resident activities occur with different frequencies at different locations within home, so the nodes’ sensing and transceiving activity differ around the home thus resulting in different residual energy. The activity context information learned by ActiSen will be embedded into the topology management and routing protocols. The following figure illustrates that sensors (in green) detect resident activities and form a context cluster. The elected cluster leader then initiates a control message to an actuator (in red) for automation assistance.
Motion sensor: Panasonic AMN 31111