This thesis aims to investigate and develop new sensing systems that can boost the widespread deployability of Internt-of-Things (IoT) and robotic-based applications. The current IoT and robotic systems are limited due to their extensive reliance on visual sensors, e.g., in bad lighting conditions. In contrast, I present my research that aims to develop accurate, reliable, and universal wireless sensing systems by leveraging advanced Machine Learning (ML) methods.
Wireless sensors provide a new source of information about the sensed physical world for these systems.
So, I present systems that provide the location, context, and privacy for wireless sensor-based IoT and robotic systems.
Firstly, I demonstrate through LocAP and MapFind how to integrate the context for physical and wireless environments -- using MapFind in tandem with the existing simultaneous localization and mapping(SLAM) algorithm to map the physical environment and using LocAP to map the wireless sensing anchors deployed in the infrastructure accurately to a few millimeters.
We further demonstrate that within this given context, using DLoc, an ML-based wireless location algorithm, can robustly (even for top 10\% corner cases) and accurately locate the user device within a few decimeters.
Finally, wireless sensors' ability to passively sense beyond visual obstacles makes many systems designed to rely on them vulnerable to security and privacy attacks. So, we developed `MIRAGE' that confuses the attacker regarding the user's actual location by providing them with a highly incorrect estimate.