Autonomous Ocean Observation: Reinforcement Learning for UAV-VLC Underwater Communication Optimization

Underwater optical data collection is strongly affected by time-varying turbidity, wave-induced misalignment, and limited onboard energy. This paper proposes a UAV-assisted visible light communication (VLC) framework driven by a scenario-specific lightweight reinforcement-learning controller for adaptive motion and transmit-power control. The control problem is formulated as a compact Markov decision process using UAV geometry, residual energy, attenuation conditions, link-quality statistics, queue status, and day/night operation. The framework also incorporates a physics-informed dual-segment optical access model covering underwater propagation, water–air interface transmission, and aerial attenuation. Simulation results on a coastal testbed show that the proposed scenario-specific lightweight DDQN controller provides a favorable coverage–energy trade-off compared with fixed-hover and greedy baselines, while remaining competitive with other learned controllers. By improving energy-aware underwater data collection, the framework supports sustainability-oriented ocean monitoring aligned with SDG 14, SDG 13, and Saudi Arabia’s RDI priority on Sustainable Environment and Supply of Essential Needs. These results support the potential of lightweight physics-informed reinforcement learning for adaptive UAV-assisted underwater sensing within the scope of a simulation-based study.