Deep Reinforcement Learning for Autonomous Underwater Navigation: A Comparative Study with DWA and Digital Twin Validation | Synapse