Smart ROUV advancements enhance Suez Canal navigation

In this section, we show how our new method outperforms state-of-the-art methods.

Comparison with baseline method

To evaluate the effectiveness of the proposed LiDAR + AI ROUV system, we compared it with two baselines: (i) Traditional sonar-only ROUV implementation⁶(ii) Representative LiDAR-only underwater mapping approaches from recent literature.⁹. This comparison focused on three metrics: (1) point cloud completeness (% of surface coverage), (2) SLAM continuity score (number of significant mapping gaps per kilometer), and (3) navigation path deviation (m), defined as the deviation of the recommended safe path from the ground truth survey line. As summarized in Table 4, our LiDAR + AI ROUV achieved the highest surface coverage (92%), the most continuous SLAM map, and the lowest average path deviation (0.35 m). In contrast, sonar-only scans suffered from high noise and low surface coverage, while LiDAR-only scans (without AI reconstruction) showed significant data gaps due to scattering. These results highlight the importance of integrating LiDAR sensing with AI-based point recovery and optimized propulsion design.

The results of this study can be summarized in terms of the potential for widespread use of laser beams in underwater surveying processes, especially in narrow, shallow and complex channels. This is shown in the form of a map in the figure, which shows the suitability of the laser and its high efficiency in penetrating different media in complex marine environments.^{twenty two}.

Plotting laser scan data with ROS data provides a clearer, more accurate, and easier-to-understand picture of the marine environment than simply using distance and angle data. The subject of this study is the Suez Canal channel, and as shown in Figure 15, out of the 166 km total length of the channel that requires investigation, six points representing three cases of cliffs that exist in the Suez Canal were selected for investigation.

The scanning process relies on a submarine diving next to the cliffs of the Suez Canal. There, a LIDAR sensor emits laser waves on the cliffs of a canal, and these waves are reflected in the form of echoes to a receiver located on the LIDAR sensor. The emitted and reflected waves are analyzed and each point is broken down into the following main information: distance and angle. This data is transmitted to the mainland, analyzed using ROS technology, and rendered in the form of SLAM. However, some points are lost due to attenuation, refraction, and interference of some laser beams. Therefore, “optimization” was carried out to predict the leading points to more accurately depict peace using machine learning, which is a method of artificial intelligence.²⁰.

The study was conducted on three types of embankments along the entire length of the Suez Canal. The first type is the armored Softbank located in the Eltina area south of Port Said. The second type is the abrasion scarp in the area of ​​Ahmed Mansi Floating Bridge and El Ferdan in southern Qantara. The third type is the steep mud cliffs found in the areas of Deversoir and Kabrit in southern Ismailia.

The first type is the armored Softbank in the Tina area, 25 kilometers south of Port Said. Its geographical location relative to the navigation channel is shown in Figure 20. In this area, the sides of the canal were made of stone due to the deterioration of the canal layer, causing coastal drift of the canal bed. Therefore, armor is considered the norm and represents the protection of these cliffs. Therefore, the cliff appears as a straight line in a cross-section of the area, as shown in Figure 20. An image of the actual cliff is shown in Figure 16. Some of the data related to LIDAR sensor scanning is shown and is transmitted from the vehicle to the mainland in the form of distance and angle. This is represented by an optimized SLAM map using Ross technology, as shown in Figure 17. In the diagram, the optimal navigation path is predicted. The data is then input into a machine learning optimization program to predict missing points, as shown in Figure 17. Based on this improvement, the route is re-estimated as shown in Figure 21.

The second type is the abrasive cliffs in the areas of Ahmed Mansi Floating Bridge and Al Fardan, located at 73.6 and 67 km south of Qantara. Its geographical location relative to the navigation channel is shown in Figure 18. This is an active coastal zone with rock deposits.^{twenty three}. As ships pass through this small bay, it is deposited there due to shallow waves created by waves colliding with each other. Therefore, these cliffs are characterized by outcrops due to pore water pressure. Also, because sand dunes are formed by the wind, rock protrusions appear on this cliff as shown in the cross section. An actual image of the cliff is shown in Figure 19. Some of the data related to the LIDAR sensor scan is shown and is transmitted from the vehicle to the mainland in the form of distance and angle. This is represented and optimized in a SLAM map using ROS technology as shown in Figure 19. The optimal navigation path is predicted as shown in the figure. The data is then fed into a machine learning optimization program to predict missing points. Based on this improvement, the route is re-estimated as shown in Figure 19.

The third type is the severe mud cliffs found in the areas of Deversoir and Kabrit in southern Ismailia. At kilometer points 77.7 and 120.8, its geographical position relative to the navigation channel is shown in Figure 20. This region is silt-rich and is influenced by many geological factors, including Coriolis flow, rip currents, and deflation. These result in Ekman transport and fetch forces that cause cliffs in some areas to become mobile muds and liquefies, like bitter lakes and foreground soils. Cliff-like dunes at the north and south entrances of Cabrit and D’Oversoir, as shown in the cross section of the area in Figure 20. The actual image of the cliff is shown in Figure 21, showing some of the data related to the scan by the LIDAR sensor. The data is transmitted from the vehicle to the mainland in the form of distance and angle, and is expressed and optimized into a SLAM map using ROS technology, as shown in Figure 21. As shown in the figure, the best route for navigation is predicted. Then input the data into a machine learning optimization program to predict missing points as shown. Based on this optimization, a path similar to Figure 21 is again assumed.

The results are output after going through a scanning process where the data is organized through an ESP 32 microcontroller and the data is sent to the mainland control station in the form of distance and angle.^{twenty four}. These are serially output on the Pont operating system, organized by the ROS system, which simulates the distance and angle to the SLAM and displays it as a map of channel erosion, giving a prediction of the path based on the initial scan data. However, the low accuracy of these maps is notable as some points are lost as a result of the waves emitted by the lidar sensor reflecting in a different direction than the lidar receiver as a result of reflections at larger angles. This creates a lack of continuity. A set of points about a position on the terrain^{twenty five}. So the data is serial and we can use the Anaconda Jupiter platform, so the data is input into programs that work with artificial intelligence, specifically machine learning. Predict points based on the data of points in 4 directions adjacent to the missing point^{twenty three}. Depending on the degree of tortuosity of the terrain in the scanned location, the usage of reviewed points can reach 50 points in each direction.²⁶.

Table 5 shows a numerical comparison of the maximum and average thrust across the tested propellers, along with the observed turbulence level and efficiency ranking. As shown in Figure 22, the bar graph provides a clear visual comparison and highlights the superior performance of the v05_1 design.

These findings justify the selection of v05_1 for the final ROUV prototype, as it provides the best balance of thrust, stability, and efficiency. By displaying thrust data in both tabular and graphical formats, results facilitate comparison and enhance the robustness of design decisions.

To ensure a structured evaluation of the proposed ROUV, a series of tests focusing on different aspects of system performance were conducted. The tests include stability under controlled conditions, disturbance rejection, SLAM mapping quality, propeller performance, and full-scale navigation trials in the Suez Canal. This structured approach ensures verification of both subsystem-level and system-level behavior. A summary is shown in Table 6.

Results from these tests show that the controller delivers robust stability, the redesigned propeller minimizes turbulence, and the AI-based SLAM reconstruction improves mapping continuity in narrow lane environments.

To position our approach vis-à-vis state-of-the-art methods, we compared the proposed LiDAR + AI ROUV with traditional sonar-based ROUVs.^6,12 and LiDAR only methods^8,9. Sonar-based approaches showed high noise and incomplete data in shallow lanes, while LiDAR-only systems produced discontinuous point clouds. In contrast, our integrated LiDAR + AI system improved the completeness, stability, and accuracy of navigation paths. The comparison is summarized in Table 7.

These results highlight that the proposed ROUV achieves higher mapping completeness and continuity with significantly lower path deviations, thereby offering clear advantages over existing approaches.