UC Irvine

The IoT revolution has provided a promising opportunity to build powerful perpetual awareness systems. Perpetual awareness systems are sensing systems characterized by continuous monitoring of spaces, people and events; they are essential to many safety and mission-critical applications, e.g. assisted living, healthcare and public safety. Today, IoT platforms are the key technology substrate for smart homes/buildings that are equipped with heterogeneous devices and diverse (often multiple) network interfaces. Data thus generated can be processed locally or at a cloud to create knowledge for diverse ubiquitous services.

Many end-to-end challenges arise in operating these deployments. The key question in this thesis is: how to ensure perpetual operations in IoT deployments? We address key challenges in perpetual smart-space applications that affect system reliability, lifetime and availability, i.e. that of energy consumption, processing overhead associated with continuous sensing, communication and processing. To understand issues associated with perpetual sensing we developed and deployed a smart assisted living platform in a senior-care facility in Montgomery County, MD to detect critical events (injury, hazardous-environment, elderly falls) that require immediate action and response, where battery-operated and wall-powered IoT devices are co-executed to ensure the safety of occupants.

We observed that diverse applications utilize data at different levels of quality; e.g. different fall detection applications utilize diverse multi-modal sensory data such as tri-axis accelerometer, video image data, .. etc. Each approach delivers different fall detection accuracy, and uses algorithms with different complexity and memory allocation, so that is why resources consumed for sensing, computation and communication vary based on the desired quality. We exploit these quality tolerances by modeling them as "space-states" and intelligently leverage the dynamic space-states to select and provision resources ( access networks, device capabilities, processing location) to reduce energy and processing overhead while ensuring application quality. Furthermore, diverse people have different levels of needs; we exploit these needs by modeling them as “personal-space-states” and leverage the dynamic workload to reduce processing overhead while ensuring an efficient perpetual IoT system.

In this thesis, we design a prototype SAFER, a novel semantic approach that utilizes context extracted, trigger actions and space-states shifts for each user to drive energy-optimized sensor/network/processing activations. To validate our approach, we derive use cases from real-world assisted living smarthomes with multiple personal and in-situ devices for targeting applications such as elderly fall detection. Through detailed testbed measurements and larger simulated scenarios, we show that our provisioning algorithms and adaptive techniques that use semantics can achieve reductions in energy dissipation, active devices and maximizing system lifetime without loss of application quality/accuracy.