Could AI Solve One of the Biggest Challenges Facing Underwater Sensors?

By Muhammad OsamaReviewed by Laura ThomsonOct 21 2026

*Important notice: This news reports on an unedited version of an accepted paper and is awaiting final editing. Therefore, the paper should not be regarded as conclusive or treated as established information. 

Keeping underwater sensor networks connected is no easy task. Ocean currents constantly shift sensors out of position, acoustic signals travel slowly, and limited battery power makes reliable communication challenging. A new study published in Scientific Reports explores how artificial intelligence (AI) could help overcome these challenges by allowing underwater sensors to work together and make smarter routing decisions.

Image Credit: Mediterranean sea .1/Shutterstock.com 

In simulations, the proposed approach extended network lifetime by up to 31% and improved packet delivery by around 10%, highlighting its potential to support more reliable and energy-efficient ocean monitoring.

Challenges in Underwater Communication

Underwater acoustic sensor networks are essential in ocean monitoring, climate research, disaster response, and coastal security. These networks rely on sound waves to transmit data over long distances, as conventional wireless signals struggle to travel through water. However, keeping underwater sensors connected comes with several challenges.

Unlike wireless communication on land, sound travels through water at about 1500 meters per second (m/s), which can cause significant data-transmission delays. Battery life is another concern, as replacing or recharging sensors in remote underwater locations can be both difficult and expensive. Ocean currents add to these challenges by constantly shifting sensors out of position, making communication links less reliable. As a result, sensors may need to use more energy to transmit data, causing some batteries to drain much faster than others.

Framework Design: CMARL and QMIX Algorithm

The researchers in the study have developed a cooperative multi-agent reinforcement learning (CMARL) routing framework based on the QMIX algorithm, where each sensor node acts as an independent agent selecting the next relay while working toward a shared network goal. The system features a Centralized Training and Decentralized Execution (CTDE) approach.

During training, a central neural network evaluates joint network information and rewards, while during operation, each node makes routing decisions based on its local observations.

Each node considers four critical factors: progress toward the surface sink, remaining battery power, available neighboring nodes, and acoustic link quality. A fully connected neural network processes these inputs through hidden layers of 128 and 64 neurons, producing 400 possible actions. Action masking excludes unreachable neighbors from consideration.

The reward system balances reliable delivery with energy efficiency. Successful transmissions receive positive rewards based on link reliability and the neighbor's remaining battery, while unnecessary retries incur penalties. If forwarding fails after three retransmissions, penalties are applied for packet loss and the burden on low-energy nodes.

The study then evaluated the framework using MATLAB software (2023a version) with its RL and Deep Learning (DL) toolboxes. Simulations covered networks of 100 to 400 nodes in a 2,000 m × 2,000 m × 2,000 m underwater volume, with 60 independent trial iterations.

Performance Evaluation and Results

Simulations showed that the QMIX-based framework outperformed the CLORP (Cross-Layer Opportunistic Routing Protocol), RLOR (Reinforcement Learning-based Opportunistic Routing), and DQIR (Deep Q-network-based Intelligent Routing). As node drift speeds increased from 1 to 3 m/s, it maintained higher packet delivery rates, averaging about 10% above CLORP. The performance gap was larger against the single-agent methods.

In a 400-node network scenario, the framework extended network lifetime by 9% compared to CLORP, 14% compared to RLOR, and 31% compared to DQIR. Lifetime was measured as the time until the first node exhausted its battery.

In simulations with 300 nodes, the network retained 1.54 times more average residual energy than CLORP and more than four times that of RLOR. It also reduced average end-to-end latency by around 18% compared to CLORP in the 300-node scenario. The Hybrid Performance Index (HPI), which combines packet delivery, energy efficiency, latency, and network lifetime, awarded the proposed framework a score of 1.0000, while CLORP scored 0.7506, RLOR scored 0.4742, and DQIR scored 0.0000. These results indicate that coordinated routing significantly improved delivery, energy distribution, and network longevity under the simulated conditions.

Click here to download a PDF copy of this page

Applications for Real-World Use

The routing framework has significant implications for long-term oceanic monitoring and tracking marine biodiversity. Sensor nodes can track tsunami precursors and tectonic activity while maintaining communication with surface buoys, reducing premature battery depletion.

In industrial and offshore sectors, this approach could support structural monitoring of underwater pipelines, drilling platforms, and seafloor cables. Autonomous underwater vehicles (AUVs) and robotic survey systems could benefit from routing that adapts to changing node positions, water depths, and ocean currents. Further testing in real marine environments will be necessary to assess its performance under operational conditions.

Future Directions for Enhanced Ocean Monitoring

The CMARL framework proved effective for enhancing routing in UASNs. By training sensor nodes collaboratively while allowing each node to make independent routing decisions, the framework improved packet delivery, reduced energy consumption, and extended network lifetime under dynamic underwater conditions.

Future work could explore additional value-decomposition techniques, the incorporation of multiple surface sinks, and more detailed ocean models. These improvements could further enhance the framework's adaptability to changing currents and network conditions, supporting long-term ocean monitoring and underwater Internet of Things deployments.

Disclaimer: The views expressed here are those of the author expressed in their private capacity and do not necessarily represent the views of AZoM.com Limited T/A AZoNetwork the owner and operator of this website. This disclaimer forms part of the Terms and conditions of use of this website.