Exploring feasibility of UAV-Based Situational Awareness Support for Maritime Autonomous Surface Ships through Scenario-Based Sea Trials

16th International Symposium on Practical Design of Ships and Other Floating Structures PRADS 2025 Ann Arbor, MI, USA, October 19th-23rd 2025 Exploring feasibility of UAV-Based Situational Awareness Support for Maritime Autonomous Surface Ships through Scenario-Based Sea Trials Dong-eon Kim1, Hyun-Jae Jo1, Tae-Jun Youn1and Geun-Tae Yim1,* 1Korea Research Institute of Ships & Ocean Engineering, Ulsan, Republic of Korea Abstract. This study investigates the feasibility of employing unmanned aerial vehicle (UAV)to enhance the situational awareness capabilities of Maritime Autonomous Surface Ships (MASS), thereby improving the reliability of their operations. However, fixed situational awareness sensors on MASS, including cameras, LiDAR, and radar, are subject to inherent physical limitations due to their static installation, reducing their effectiveness in dynamic and complex navigational environments. To address these limitations, data collected from scenario-based real-sea trials were applied in simulations to evaluate the effectiveness of UAV-assisted situational awareness. For this purpose, we designed and conducted trials using the sea-trial testbed vessel HaeyangNuri. The representative operational scenario was set in a geographically constrained channel. For the simulation, the vessel’s motion was modeled using the Velocity Obstacle method, and a weighted exponential model, incorporating the distance and time to the closest point of approach (DCPA and TCPA), was developed to quantitatively assess collision risk. The results demonstrated reduced collision risk and improved operational stability when UAV were employed. These findings highlight the potential of UAV to support MASS navigation, and future work will focus on developing a robust MASS–UAV cooperative framework for broad operational applications to further enhance the operational reliability of MASS. Key words: Maritime Autonomous Surface Ship, Sea trial, Simulation, Situational awareness, Unmanned Aerial Vehicle 1. Introduction Recent advances in artificial intelligence (AI), data communications technology, and sensing hardware have accelerated research and development on the automation of various mobile platforms. In the maritime sector, several major initiatives, such as KASS(Korea Autonomous Surface Ship)project in Korea, One Sea Association in Finland, Meguri 2040 in Japan among others, are pursuing the commercialization of Autonomous Maritime Surface Ship (MASS). According to the International Maritime Organization (IMO), a MASS is a vessel that is capable of operating independently without human intervention, although remote operation may be required depending on the level of autonomy. Given that conventional ship operations require a considerable workforce of specialized personnel, and that approximately 85% of maritime accidents are reported to be caused by human error, MASS technology is expected to deliver significant gains in both efficiency and safety[1]. The situational awareness system is considered a core technology of MASS, designed to support the detection and recognition of surrounding objects through sensors and to assist in avoiding hazardous situations. To facilitate complementary situational awareness, it typically integrates multi-sensor data, including RADAR, LiDAR, AIS, and cameras.[2]By processing and fusing real-time information, the system can contribute to safer navigation, collision avoidance, and more informed decision-making. The systems have been developed and demonstrated in various MASS technology development projects, and there are also cases where they have been developed and commercialized as auxiliary systems for conventional ships. Within the KASS Project, the intelligent Situational Awareness System (iSAS)has been developed[3]. iSAS uses sensors mounted on the ship and processes raw sensor data with deep-learning-based detection algorithms to generate real-time object recognition and status estimation. It is installed on the 25-meter *Correspondence to: [email protected] 1 testbed vessel HaeyangNuri, which has been established and operated for the verification and validation (V&V)of MASS technologies. Using this platform, real-sea trials have been conducted to evaluate the system’s object detection performance, collect operational data, and perform integrated trials in conjunction with the autonomous navigation system.[4][5] In international navigation, vessels may encounter situations where shipboard sensors, constrained by their fixed installation, exhibit inherent limitations in detection coverage. These limitations are particularly evident in areas with geographical constraints, such as narrow channels, breakwaters, or nearby islands. Given that such conditions are known to pose elevated accident risks for conventional vessels[6][7], the International Maritime Organization (IMO)has mandated the use of the Automatic Identification System (AIS)—a system that broadcasts real-time vessel identity, position, and course using radio waves—and established regulations for the deployment and operation of Vessel Traffic Services (VTS)in high-risk maritime areas. However, despite these measures, maritime accidents continue to be reported, highlighting the inherent limitations of VTS—its reliance on human operators and passive data sources such as radar, AIS, and voluntary ship reporting—and raising persistent concerns about the reliability and integrity of AIS data[8][9]. Although extensive research is underway to address the limitations of both VTS and AIS[10] [11], a MASS must be capable of independently ensuring robust situational awareness to operate safely in diverse and dynamic environments without relying solely on external aids. This independence is particularly critical for MASS, as its data-driven decision-making architecture can be highly vulnerable when dependent on potentially unreliable external inputs. To help address the limitations of fixed shipboard sensors, the use of Unmanned Aerial Vehicle (UAV) as active and mobile situational awareness platforms has been suggested as a potential approach. UAV, with their mobility, may enhance situational awareness beyond the capabilities of fixed shipboard sensors and contribute to improved operational safety of MASS. Multi-domain operational cases integrating land and air systems have been reported, such as military operations, logistics support, and autonomous vehicle assistance. In the maritime domain, various independent studies have been conducted on the use of UAV, such as inspection of ship hulls and tanks, delivery of supplies to warship[12], UAV landing technologies under dynamic sea conditions[13]. Research on cooperative USV–UAV systems for navigation support or man-over-board searches by using image-based object recognition and distance estimation has also been actively pursued[14][15][16]. However, research on the use of UAV to improve the safe operation of MASS remains in its early stages[17]. Based on this background, this study performs simulation using data from sea trial conducted with the HaeyangNuri to examine the feasibility of utilizing UAV to enhance the situational awareness capabilities of MASS. Section 2 introduces the platforms used in the sea trials. Section 3 describes the methodology, including the applied algorithms and analytical framework. Section 4 discusses the results, draws conclusions, and outlines directions for future research. 2. Experimental Platforms for Real-Sea Trial The Haeyang Nuri and a commercially available UAV were utilized in this study. The Haeyang Nuri, built under the KASS project, serves as a sea-trial testbed vessel and is actively used for the demonstration and performance validation of MASS technologies under real-sea conditions. A flat, unobstructed surface is essential for stable UAV take-off and landing. However, the upper strucFigure 1. The HaeyangNuri, a sea-trial testbed vessel 2 Figure 2. UAV deck & sensors installed on HaeyangNuri ture of the Haeyang Nuri is densely fitted with sensors and antennas, creating spatial constraints for UAV deployment. To address this, a 4 m × 4 m UAV deck was installed on the vessel’s top deck. In line with CAP 437: Standards for Offshore Helicopter Landing Areas, the deck meets the recommended standard for rotor diameters up to 3.2 m, based on the 1.25× safety margin[18]. The deck is constructed with FRP–PVC sandwich panels for weight reduction and structural strength, finished with anti-slip, damping materials for safety. A clearance at the hull–deck junction improves airflow and reduces the influence of relative wind during navigation. 2.1. Data acquired by sea-trial A scenario was developed to reflect navigational situations commonly experienced by vessels engaged in international trade, particularly in geographically constrained environments such as channel approaches to the Port. In such waters, obstructions—such as headlands, islands, industrial facilities, and breakwaters— can restrict the line of sight of vessels, which may result in limited visibility, delayed recognition of potential collision risks, and increased threats to safe navigation. To replicate these conditions, a scaled-down sea trial was conducted at coastal site in Ulsan, Republic of Korea, under WMO Sea State Code 2 (Smooth)conditions, with significant wave heights below 0.5 m and wind speeds not exceeding 8 knots. The HaeyangNuri and the UAV were manually controlled as per scenario by a crew and a UAV operator, respectively. Cameras mounted on the HaeyangNuri and the UAV used in the trial were commercial products. Motion data of both the own ship and target vessel, as well as detection results from fixed shipboard sensors and UAV, were collected for use in the simulation. Figure 3 provides an overview of the sea-trial process. The trial site was situated near the breakwater of Bangojin Port (approx. 35°28’38”N, 129°25’33”E), an area characterized by terrain-induced line-of-sight limitations. Figure 4(a)presents the Electronic Chart Display and Information System (ECDIS)view of the Haeyang Nuri approaching the breakwater before Figure 3. Overview of the scenario 3 Figure 4. Still image of trial of Scenario; (a)ECDIS, (b)Mounted cameras,(c)UAV entering the port, where the triangular marker indicates the target ship’s position based on AIS signals. Figure 4(b)shows a still image from the vessel-mounted cameras, while Figure 4(c)provides a UAV-captured image. 2.2. Simulation design In the simulation, the dynamic motion of the vessels was modeled using the Velocity Obstacle (VO) algorithm[19]. For the own ship, the motion characteristics of the Haeyang Nuri were incorporated to reflect its maneuvering behavior. The initial positions and speeds of both the own ship and the target ship were reproduced from data collected during the sea trial, and the simulation environment was designed to replicate the conditions of the trial site. To evaluate detection performance, visual data acquired from both the vessel-mounted cameras and the UAV were processed using YOLO11[20], an open-source AI-based object detection model. A time delay was observed between the two platforms: while the UAV successfully detected the target ship first, the fixed camera required approximately 110 seconds longer to achieve detection under the same conditions. To quantify the collision risk (R)between the own ship and the target ship, a weighted exponential function model considering both the Distance to Closest Point of Approach (DCPA)and the Time to Closest Point of Approach (TCPA)was employed. The risk index ranges from 0 to 1, where values approaching 1 indicate higher risk. The index increases only when both distance and time simultaneously indicate elevated collision risk. R=e−DCPA d0×e−TCPA t0(1) where d0is the reference DCPA and t0is the reference TCPA. Figure 5. Exponential decay of the R: (a)TCPA variation for fixed DCPA values, and (b)DCPA variation for fixed TCPA values 4 Figure 6. Comparison of the own ship’s trajectory and collision risk index: (a)fixed-camera case, (b)UAV-assisted case. The dashed line is the intended path, and the red line is the actual track. 2.3. Result The simulation results demonstrate a significant performance improvement in collision-avoidance behavior when utilizing UAV-assisted detection compared to the fixed camera setup. As shown in Table 1, the UAV achieved an earlier detection time of 110 s, compared to the fixed camera. This earlier detection led to a 27.8% reduction in the maximum risk index (from 0.4827 to 0.3485)and a 14.4% decrease in the variance of the risk index. The standard deviation also decreased by 7.2%, indicating more stable collision-avoidance responses during the simulation. In terms of navigational accuracy, the track error—defined as the accumulated distance deviating from the intended path—was reduced from 383 m in the fixed camera case to 258 m in the UAV-assisted case, representing a 32.6% improvement. The reduction in the maximum risk index suggests that the likelihood of a potential collision was effectively lowered when UAV data were incorporated. Similarly, the decreases in variance and standard deviation reflect more consistent and predictable collision-avoidance behavior, minimizing abrupt changes in risk levels. The track error improvement further indicates that the vessel was able to maintain a trajectory closer to its intended path, reducing unnecessary deviations during maneuvering. Table 1. Comparison of collision-avoidance performance between fixed camera and UAV-assisted detection Metric Fixed camera UAV Improvement Max. risk index 0.4827 0.3485 −27.8% Variance 0.0097 0.0083 −14.4% Standard deviation 0.0984 0.0913 −7.2% Track error [m]383 258 −32.6% Note: The fixed camera exhibited a detection delay of approximately 110 seconds compared to the UAV. 5 3. Discussion & further work The results of this study demonstrate that UAV-assisted detection significantly enhances the timeliness of situational awareness, particularly in environments with line-of-sight constraints. The reduction in detection time directly contributed to a lower maximum risk index and reduced variability, indicating more predictable and stable collision-avoidance behavior. Furthermore, the decrease in track error suggests improved trajectory adherence, which is critical for safe navigation in constrained waterways where maneuvering margins are limited. Despite these promising results, several limitations remain. The sea-trial data used for the simulation were derived from a scaled-down test environment, and further evaluations are required to verify the operational stability of UAVs under diverse maritime conditions, including high winds, precipitation, and saltwater exposure. Moreover, the current evaluation focused on a single navigational scenario, which may not fully capture the range of conditions encountered in global maritime operations. Building on these findings, a MASS–UAV cooperation system is expected to further enhance the operational reliability of MASS by providing third-person perspectives to remote operators during navigation in high-traffic areas, supporting local route optimization, security monitoring, and safety management. Building on the prototype developed and demonstrated with the Haeyang Nuri, future research will broaden the range of operational scenarios and refine this framework into a robust, fully integrated MASS–UAV cooperation system for practical application in diverse and dynamic maritime environments. Acknowledgments This research was supported by Korea Research Institute of Ships and Ocean engineering a grant from Endowment Project of “A study on sea demonstration of aerial drone operation as auxiliary system of situational awareness for autonomous ships”funded by Ministry of Oceans and Fisheries(2520000694) References [1]Jin Kim and Hwasup Jang. Trends and preparations for maritime autonomous surface ship technology(자율운항 선박 기술동향 및 준비). Bulletin of the society of naval architects of korea,56(4):4–7, 2019. [2]Les Elkins, Drew Sellers, and W. Reynolds Monach. The autonomous maritime navigation (amn)project: Field tests, autonomous and cooperative behaviors, data fusion, sensors, and vehicles. Journal of Field Robotics, 27(6):790–818, 2010. [3]Hyun Taek Choi, Jeonghong Park, Jinwoo Choi, Minju Kang, Yeongjun Lee, Jongdae Jung, Jonghwi Kim, Hyeokjun Kweon, Jinwhan Kim, Kuk Jin Yoon, et al. Design and preliminary results of novel situational awareness system for autonomous ship based on artificial intelligence techniques. Journal of Institute of Control, Robotics and Systems, 27(8):556–564, 2021. [4]Minju Kang, Kibeom Choo, Jinwoo Choi, Hyun-Taek Choi, Hye jin Kim, and Jeonghong Park. A multimodal dataset for maritime object detection. In Proceedings of the Korean Society of Mechanical Engineers Spring and Autumn Conference, Jeju, South Korea, 04 2025. [5]Kibeom Choo, Jeonghong Park, Jinwoo Choi, Minju Kang, and Hyun-Taek Choi. Current status of situational awareness system demonstrations for collision and accident prevention in mass(자율운항선박 충돌 및 사고방지 상황인식 시스템 실증 현황). Proceedings of the Korean Institute of Navigation and Port Research Conference, 2024(11):43–44, 2024. [6]N. Endrina, J. C. Rasero, and J. Montewka. Risk analysis for maritime traffic in the strait of gibraltar and improvement proposal. In C. Guedes Soares and Angelo P. Teixeira, editors, Maritime Transportation and Harvesting of Sea Resources, volume 1, pages 223–231, United States, January 2016. CRC Press. International Congress of the International Maritime Association of the Mediterranean, IMAM ; Conference date: 09-10-2017 Through 11-10-2017. [7]S Yildiz, F Tonoğlu, Ö Uğurlu, S Loughney, and J Wang. Spatial and statistical analysis of operational conditions contributing to marine accidents in the singapore strait. Journal of Marine Science and Engineering, 10(12), December 2022. 6 [8]Andrej Androjna and Marko Perkovič. Ais data falsification - how long will it be before we can no longer trustais? In 2024 IEEE International Workshop on Metrology for the Sea; Learning to Measure Sea Health Parameters (MetroSea), pages 301–306, 2024. [9]Windward Ltd. Gps jamming falsely placed vlcc front eagle in iran prior to collision, June 2025. Accessed: 2025-08-31. [10]Xing Wu, Afifa Rahman, and Victor A. Zaloom. Study of travel behavior of vessels in narrow waterways using ais data – a case study in sabine-neches waterways. Ocean Engineering, 147:399–413, 2018. [11]Lei Zhang, Jiahao Ge, Floris Goerlandt, Lei Du, Tuowei Chen, Tingting Gu, Langxiong Gan, and Xiaobin Li. A method for enhancing the traffic situation awareness of vessel traffic service operators by identifying high risk ships in complex navigation conditions. Journal of Marine Science and Engineering, 13:379, 02 2025. [12]Kai Greet. Drones to deliver cargo during royal navy carrier strike group 25 deployment. https:// theaviationist.com/2025/04/07/csg25-drone-cargo-delivery/, April 2025. Accessed: 2025-05-13. [13]Shadi Abujoub, Johanna McPhee, and Rishad A. Irani. Methodologies for landing autonomous aerial vehicles on maritime vessels. Aerospace Science and Technology, 106:106169, 2020. [14]Chen Cheng, Dong Liu, Jin-Hui Du, and Yong-Zheng Li. Research on visual perception for coordinated air–sea through a cooperative usv-uav system. Journal of Marine Science and Engineering, 11(10), 2023. [15]Yuanda Wang, Wenzhang Liu, Jian Liu, and Changyin Sun. Cooperative usv–uav marine search and rescue with visual navigation and reinforcement learning-based control. ISA Transactions, 137:222–235, 2023. [16]Xuesu Xiao, Jan Dufek, Tim Woodbury, and Robin Murphy. Uav assisted usv visual navigation for marine mass casualty incident response. In 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 6105–6110, 2017. [17]D.-K. Kim, J.-H. Kim, H.-H. Son, S.-W. Choi, D.-H. Kim, C. Y. Yeo, and J.-Y. Park. A study on a real-time aerial image-based uav-usv cooperative guidance and control algorithm. Journal of the Society of Naval Architects of Korea, 61(5):324–335, 2024. Published October 20, 2024. [18]UK Civil Aviation Authority. CAP 437: Standards for Offshore Helicopter Landing Areas. Civil Aviation Authority, Aviation House, Gatwick Airport South, West Sussex, RH6 0YR, UK, 9 edition, February 2023. Available at: https://www.caa.co.uk/CAP437. [19]P. Fiorini and Z. Shiller. Motion planning in dynamic environments using the relative velocity paradigm. In [1993] Proceedings IEEE International Conference on Robotics and Automation, pages 560–565 vol.1, 1993. [20]Ultralytics. Yolo11: Real-time object detection, 2024. Accessed: 2025-08-31. 7