Despite covering 71% of Earth's surface, 80% of the oceans remain unexplored and unmapped. Even if extremely important, ocean exploration is just one of many applications that could benefit from networks of connected underwater devices capable of sensing the environment and communicating with each other wirelessly. Other applications that could benefit from underwater wireless networks include disaster prevention, oil and gas pipeline monitoring, aquaculture, habitat restoration, and national security.Networks of underwater wireless devices typically use acoustic waves for communicating as that is the sole technology capable of propagating for long distances in water. However, these networks encounter several obstacles, such as slow propagation speed, the Doppler effect, reflections, and constantly changing channel characteristics.The basic building block of underwater acoustic wireless networks is the underwater acoustic wireless modem. Over time, attempts have been made to create faster and more reliable underwater acoustic modems, such as those manufactured by Teledyne Benthos and Evologics. Nevertheless, the acoustic modems currently available in the market suffer from two primary issues: slow data rates and an inflexible software and hardware architecture which impedes new protocol and standard development.Intending to advance the field of underwater wireless networks, i.e., realizing networks that are faster than the state of the art and that can adapt to the ever-changing conditions of the marine environment, this work investigates the field of Medium Access Control protocols in underwater networks. Specifically, it focuses on MAC protocols that rely on Orthogonal Frequency-Division Multiplexing (OFDM) physical layers, the physical layer technique that is also used in Wi-Fi and LTE. To investigate OFDM-based MAC layers in the real world and at scale, we follow two parallel xii paths: i) for the real-world MAC layer testing we first design, develop, and test the SEANet modem, a software-defined acoustic platform. We then design, implement, and test OFDM-based MAC protocols using a small network of SEANet modems. ii) To test the efficacy of OFDM-based MAC layer protocols in large networks we design implement and test several OFDM-based MAC Protocols on multi-hop networks using the DESERT simulator. The MAC protocols tested on the DESERT simulator leverage the OFDM technology at the physical layer as well as reinforcement learning.Experiments carried out in the simulated environment and the real network of SEANet modems confirm that OFDM-based MAC Protocols achieve higher data rates with respect to the state of the art and allow the network to be smarter and more reliable, especially when selective interference is present. The addition of reinforcement learning to the OFDM-based MAC protocols produces networks that are adaptable with minimal human contribution.xiii