In this paper, we propose a Medium Access Control (MAC) protocol for mobile Underwater Acoustic Sensor Networks (UWASNs), based on framed-ALOHA and Reinforcement Learning (RL) principles. The protocol capitalizes on the large propagation delays in underwater acoustic channels to enhance network throughput. Deployed sensor nodes function as RL agents, autonomously learning optimal transmission strategies concerning both time slot allocation and packet transmission delays. This approach eliminates the need for gathering detailed information about the network topology and strengthens the network's resilience and adaptability. To evaluate its performance, we conducted UWASN simulations with different node mobility patterns. The simulation results show that the proposed protocol outperforms the existing RL-based UW-ALOHA-Q solution, achieving up to 50.7% improvement in channel utilization under the most challenging conditions, while sustaining a high level of fairness.