Most research in underwater acoustic communication deals with the physical layer, with primitive and inflexible multiple access, despite the medium access control layer playing a crucial role in providing efficient communication. The underwater propagation environment is complex and characterised by highly variable spatial and temporal conditions due to extensive multi-path phenomena, space variability (e.g. shadow zones), fast time-variability of the channel and high Doppler spread. This PhD will explore the potential for MAC layer adaptation based on machine learning. Simulation based modelling and performance evaluation will be undertaken, backed up by mathematical analysis, for typical scenarios. Distributed protocols will be designed which can adapt the use of resources in frequency and/or time, to avoid interference and in response to varying link availability.