UC Irvine

Improvements in Internet connectivity and advances in smart personal devices have enabled the rise of the Internet of Things (IoT) in real-world communities.

Community IoT deployments utilize low-cost devices, often deployed in-situ in a relatively stable environment, to create real-time situation awareness.

Our experience in operating and maintaining prototype IoT systems in real-world testbeds indicates that integrating mobile devices with in-situ platforms is a promising approach to increase the reliability and sustainability of commonplace community IoT applications.

In particular, mobile devices can be leveraged to compensate for the non-uniform availability of infrastructure efficiently. Realizing the potential of the combined ``mobile and in-situ'' deployments requires us to address a new set of challenges for data collection in dynamic settings.

In this thesis, we propose planning-based approaches to the efficient operation and maintenance of community-scale IoT deployments that consist of both mobile and in-situ devices.

Our proposed techniques leverage the prior knowledge of data characteristics, device heterogeneity, community infrastructure, and application needs.

The goal is to optimize the activities of the devices under data budgets and timeliness constraints and seek a balance between data utility (i.e., accuracy, importance, and timeliness) and operational cost.

We explore our solution within the context of urban environmental sensing and address three major research problems regarding IoT data generation, data upload, and sensor calibration (i.e., maintenance), respectively.

First, we propose a spatiotemporal scheduling framework that regulates the data generation activities of participating devices. The framework employs online planning algorithms that optimize the spatiotemporal coverage of collected data to meet the application requirements of heterogeneous data types.

Second, in the case of non-uniform network availability, we design a two-phase upload planning approach that creates data upload plans (i.e., when, where, and what to upload) for mobile data collectors before their departure (i.e., the static planning phase); the plan can then be adjusted during execution based on the dynamicity they observe (i.e., the dynamic adaptation phase).

Finally, to increase accuracy and ensure consistency of collected data during a relatively long period of operation, we propose a multi-sensor calibration planning solution that determines the number of calibration iterations, the time at which they take place, and for each of them, the sensors to calibrate and the number and paths of mobile calibrators.

Together, the proposed techniques provide a comprehensive approach to generate intelligent plans for data collection and sensor maintenance in smart communities that can fully exploit the capabilities of mobile and in-situ devices.

We validate our approach in a proof-of-concept IoT system, SCALECycle (based on the SCALE affordable IoT solution for communities), and conduct measurement studies at the community scale.