Collecting, Analyzing, and Using Fine-Grain Sensor Data with Mobile Platforms

We investigate the challenges of collecting sensor measurements with mobile platforms to reveal spatio-temporal insights across our surrounding environments and enable novel data-driven algorithms. Our efforts aim to take advantage of the increasing ubiquity of mobile platforms are already equipped with sensors. Today, sensor measurements are primarily used in limited settings (outdoor GPS) with coarsegrain accuracy (geofencing). The goal of this thesis is to investigate developments towards fine-grain sensor data with centimeter-level accuracy. In particular, we show that fine-grain sensor maps enable novel data-driven algorithms that leverage awareness of surrounding conditions. In our experience, the biggest challenge is not in using sensor data but actually ensuring continuously up-to-date sensor maps. Therefore, this thesis seeks to addresses the key aspects of collecting, analyzing, and using fine-grain sensor measurements. Mobile platforms have a distinct advantage over existing data collection efforts due to recent developments in localization and navigation that allow a device to capture labeled sensor data. Benefits include lower costs due to reduced dedicated sensor hardware requirements, reduced human effort due to automatic localization, and increased measurement diversity from moving around. In this thesis, our contributions focus specifically on wireless and indoor climate sensors, where we contend with different measurements characteristics. For data collection, we present a data collection framework for both autonomous robots and cell phones that records data from available sensors. We also develop two algorithms to generate paths for mobile platforms to execute that prioritize sensor measurement needs. For data analysis, we present a discrete spatio-temporal representation so that we can extract finegrain insights. We specifically introduce navigation adjusted grids, which performs spatial decomposition over the reachable workspace. For data usage, we develop data-driven wireless handoff algorithms that use fine-grain wireless maps to ensure accurate and timely wireless handoff decisions. We demonstrate the value of these efforts with several concrete contributions. First, we use mobile robots to gather sensor measurements over several months across two enterprise environments. Through our analysis efforts, we reveal finegrain spatio-temporal insights that show wireless conditions (wireless access point coverage and channel allocation) and indoor climate variations (temperature and humidity changes due to HVAC policies). We also show that data-driven wireless handoffs are able to significantly improve wireless performance for our robot that originally faced intermittent wireless connectivity issues while moving. With the increasing pervasiveness and technological developments of mobile devices, this thesis investigates the opportunity to leverage their sensor hardware to better integrate computing technology with our surrounding environments.