Collaborative, Intelligent, and Adaptive Systems for the Low-Power Internet of Things

Doctoral thesis, 2022

With the emergence of the Internet of Things (IoT), more and more devices are getting equipped with communication capabilities, often via wireless radios. Their deployments pave the way for new and mission-critical applications: cars will communicate with nearby vehicles to coordinate at intersections; industrial wireless closed-loop systems will improve operational safety in factories; while swarms of drones will coordinate to plan collision-free trajectories. To achieve these goals, IoT devices will need to communicate, coordinate, and collaborate over the wireless medium. However, these envisioned applications necessitate new characteristics that current solutions and protocols cannot fulfill: IoT devices require consistency guarantees from their communication and demand for adaptive behavior in complex and dynamic environments.

In this thesis, we design, implement, and evaluate systems and mechanisms to enable safe coordination and adaptivity for the smallest IoT devices. To ensure consistent coordination, we bring fault-tolerant consensus to low-power wireless communication and introduce Wireless Paxos, a flavor of the Paxos algorithm specifically tailored to low-power IoT. We then present STARC, a wireless coordination mechanism for intersection management combining commit semantics with synchronous transmissions. To enable adaptivity in the wireless networking stack, we introduce Dimmer and eAFH. Dimmer combines Reinforcement Learning and Multi-Armed Bandits to adapt its communication parameters and counteract the adverse effects of wireless interference at runtime while optimizing energy consumption in normal conditions. eAFH provides dynamic channel management in Bluetooth Low Energy by excluding and dynamically re-including channels in scenarios with mobility. Finally, we demonstrate with BlueSeer that a device can classify its environment, i.e., recognize whether it is located in a home, office, street, or transport, solely from received Bluetooth Low Energy signals fed into an embedded machine learning model. BlueSeer therefore increases the intelligence of the smallest IoT devices, allowing them to adapt their behaviors to their current surroundings.