Exploring Novel Sensing Paradigms for Existing Wireless Infrastructure

Over the past few decades, we have experienced the permeation of commodity wireless devices into every facet of our existence: our lives at home, our commutes and community infrastructure, and our workplaces. Wireless devices are already on board many sensing platforms, thus making them appealing to explore for adding sensing capabilities without increasing bulk. For example, consider the ability to use WiFi, already aboard a UAV for transmitting video data, to additionally provide information on the environment. In this dissertation, we explore opportunities in re-engineering these already widely deployed technologies (specifically, WiFi, mmWave RADAR, and cameras) for sensing object material and long-range depth imaging – challenges that traditionally require large additional sensor loads (e.g., X-band and K-band RADAR) and struggle with occlusions like fog and walls (in the case of cameras and LIDAR). The ability to sense object material and image depth, in both line-of-sight and non-light-of-sight settings,

would enable future vehicles and cities to better respond to environmental obstacles. For example, we envision a future disaster-response drone that can navigate a scene with responses guided by awareness of occlusion type – e.g., detecting a tree and flying lower to avoid getting caught in branches, or detecting a human under rubble sending a distress call. Beyond object type, the ability to image at a long range is also helpful for responding quickly to objects in the environment, reducing the flight path (and thus,

battery) necessary to survey a scene, and providing environment dimensions to better guide rescue crews

on equipment needs. We approach this vision first from the perspective of exploring novel sensing capabilities of WiFi, since it is comparatively less studied than RADAR and vision. Specifically, we develop and evaluate a WiFibased

approach to both localize and identify object type (specifically composing material) of obstacles in the environment in our system IntuWition. IntuWition uses polarimetry – the measurement of how the polarization of signals is changed as it passes through or reflects off objects – as a position- and orientationrobust

way to measure the material type of objects in the environment. This leaves the challenge of imaging objects in a scene, ideally at a long range and at high resolutions. mmWave RADAR and cameras emerge as the best candidates for this, but they individually face some challenges: mmWave RADAR has extremely poor angular resolution (about 1 degree in azimuth and about 11 degrees in elevation for a top-of-the-line model), while cameras have poor depth resolution, particularly at long ranges. We present the development and evaluation of a hybrid mmWave/camera system, Metamoran, that uses camera

information to guide the de-noising, angular positioning, detection, and imaging by a mmWave RADAR. These approaches help to demonstrate that new approaches to sensing that exploit the proliferation of wireless devices can enable increased environmental awareness.