Abstract

Modern Intelligent Transportation Systems (ITSs) employ communication technologies in order to ameliorate the passenger's commuting experience. Vehicular Networking lies at the core of inaugurating an efficient transportation system and aims at transforming vehicles into smart mobile entities that are able to sense their surroundings, collect information about the environment and communicate with each other as well as with Roadside Units (RSUs) deployed alongside roadways. As such, the novel communication paradigm of vehicular networking gave birth to an ITS that embraces a wide variety of applications including but not limited to: traffic management, passenger and road safety, environment monitoring and road surveillance, hot-spot guidance, Drive Thru Internet access, remote region connectivity, and so forth. Furthermore, with the rapid development of computation and communication technologies, the Internet of Vehicles (IoV) promises huge commercial interest and research value, thereby attracting a significant industrial and academic attention.

This thesis studies and analyses fundamentally challenging problems in the context of vehicular environments and proposes new techniques targeting the improvement of the performance of ITSs envisioned to play a remarkable role in the IoV era. Unlike existing wireless mobile networks, vehicular networks possess unique characteristics, including high node mobility and a rapidly-changing topology, which should be carefully accounted for. Four major problems from the pool of existing vehicular networking persisting challenges will be addressed in this thesis, namely: a) establishing a connectivity path in a highly dynamic Vehicular Ad Hoc Network, b) examining the performance of Vehicle-to-Infrastructure communication Medium Access Control schemes, c) addressing the scheduling problem of a vehicular networking scenario encompassing an energy-limited RSU by exploiting machine learning techniques, particularly reinforcement learning, to train an agent to make appropriate decisions and develop a scheduling policy that prolongs the network's operational status and allows for acceptable Quality-of-Service levels and d) overcoming the limitations of reinforcement learning techniques in high-dimensional input scenarios by exploiting recent advances in deep learning in an effort to satisfy the driver's well-being as well as his demand for continuous connectivity in a green, balanced, connected and efficient vehicular network. These problems will be extensively studied throughout this thesis, followed by discussions that highlight open research directions worth further investigations.